PREDICTIVE ANALYSIS OF POWER

GENERATION FROM FOOTSTEPS

USING MACHINE LEARNING
A
Project Report submitted to Sandip University Nashik

In partial Fulfillment for the awards of Degree of Engineering in

Computer Science and Engineering

Submitted by
Mr Hemant P. Chauthe 220105232008
Mr Dhananjay D. Patil 220105232010
Ms Harshada P. Pawar 220105232007
Ms Revati M. Jadhav 220105232019

Under the Guidance of

Dr. Supriya Bhosale
PROF. , SOCSE, SANDIP UNIVERSITY, NASHIK

May, 2024-25

Department of Computer Science and Engineering

School of Computer Sciences and Engineering
Sandip University Nashik
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Sandip University Nashik

School of Computer Sciences and Engineering
Department of Computer Science and Engineering
(2024-25)

Certificate
This is to certify that,
Mr Hemant P. Chauthe, 220105232008
Mr Dhananjay D. Patil, 220105232010
Ms Harshada P. Pawar, 220105232007
Ms Revati M. Jadhav, 220105232019
have successfully completed the project entitled “ Predictive Analysis Of Power Generation
From Footsteps Using Machine Learning ”, under my guidance in partial fulfillment of
the requirements for the degree of Bachelor of Engineering in C om puer S ci enc e and
Engi ne eri n g under t he S andi p University Nashik during the academic year 2024-25.

Date : ……………….
Place:……………….

……………………………….
Dr. Supriya Bhosale
Project Guide

………………………… ……………………….
Dr. Umesh Pawar Dr. Pawan Bhaladhare
Head, Dean
Department of Computer Science and Engg

Examiner :………………………………………………………………………………….

ii
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Acknowledgements

We would like to extend our heartfelt gratitude to our parents, project guide, project co-
ordinator, head of department, dean, and all other university staff for their unwavering
support throughout this project. Their assistance was invaluable in helping us to complete our
final year degree project, which contributed significantly to our Bachelor of Technology in
Computer Science and Engineering. We truly appreciated the interest and learning
experiences that accompanied this endeavor.

We express our deepest appreciation to our Project Guide, Dr. Supriya Bhosale, for her
expert guidance and technical assistance throughout the project work, which is helpful to
learn from this project and implement essential strategic plans for the successful completion
and implementation of this project. Her technical assistance helped us to learn technical
aspects of software project management and helped us to gain practical knowledge from this
project.

We would like to extend our heartfelt gratitude to our Project Co-ordinator, Prof. Abdul
Kayyum Shaikh Sir, for his support and for giving important instructions to complete this
project throughout the final year. His guidance about project guidelines and project structure
helped us to complete the project within the timeline using our true potential.

We express our deepest appreciation to the Head of Department, Dr. Umesh Pawar Sir,
for arranging different technical sessions based on project development and project-related
activities, which helped us to plan our project development activities according to industry
standards.

We would like to extend our heartfelt gratitude to the Dean, Dr. Pawan Bhaladhare Sir,
for providing all the necessary college resources and assigning different faculty, who have
expertise in different project domains and have knowledge of the technical aspects of
different project domains.

We must express sincere, heartfelt gratitude towards Sandip University for providing us with
all the necessary academic resources and infrastructure; without it, we could not have
completed this final year degree project in the award of Bachelor of Technology in Computer

Mr. Hemant P. Chauthe

Mr. Dhananjay D. Patil
Ms. Harshada P. Pawar
Ms. Revati M. Jadhav

iii
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Abstract

Energy has long been the pulse of human civilization, fueling progress and innovation. Yet, as
conventional sources dwindle, the search for sustainable alternatives becomes not just a
necessity but an inevitability. This project explores a concept hidden in plain sight harnessing
the energy of footsteps. Every step carries kinetic potential, an untapped rhythm of daily life
waiting to be transformed.

Piezoelectric tiles embedded in high-footfall areas homes, schools, railway stations capture
this overlooked energy, converting motion into electricity. Each step, no longer just a
movement, becomes a contribution to a self-sustaining cycle, where human activity fuels itself.
But beyond the conversion of energy, intelligence refines its efficiency. Machine learning
breathes adaptability into this system. By employing Linear Regression and Random Forest
Regression, the technology does not merely generate power it learns. Real-time data analysis
predicts and optimizes energy output, creating an evolving system that grows smarter with
every footstep. The results unfold on an interactive Python-powered dashboard, where raw
motion turns into visualized potential.

In a world that never stands still, where crowded streets and bustling transit hubs define the
pulse of modern life, this idea aligns seamlessly with the natural flow of human movement.
Especially in a country like India, where millions move in unison every day, the potential is
immense. This is more than an experiment in alternative energy it is a vision of sustainability
woven into the fabric of urban existence. One step, one spark, one future powered by motion
itself.

Keywords: Footstep Energy Harvesting, Machine Learning, Predictive Analysis, Sustainable

Energy, Smart Cities, Piezoelectric Sensors.

iv
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Table of Contents
1 Introduction.....................................................................................................1- 4
1.1 Overview.................................................................................1
1.2 Brief Description.....................................................................1
1.3 Problem Definition..................................................................2
1.4 Objective.................................................................................3
1.5 Organization of Report............................................................3
2 Literature Survey ..........................................................................................5 - 9
Related work done in the previous papers with their advantages and disadvantages.
3 Software Requirements Specification .........................................................10 - 39
3.1 Introduction..............................................................................10
3.1.1 Purpose......................................................................10
3.1.2 Project Scope.............................................................10
3.1.3 Design and Implementation Constrains.....................11
3.1.4 Assumptions and Dependencies.................................13
3.2 System Features (Use Case Diagrams).....................................15
3.2.1System Feature 1(Functional Requirement)................15
3.2.2System Feature2 (Functional Requirement).................17
3.3 External Interface Requirements................................................18
3.3.1User Interfaces..............................................................18
3.3.2 Hardware Interfaces.....................................................23
3.3.3 Software Interfaces.......................................................25
3.3.4 Communication Interfaces...........................................27
3.4 Nonfunctional Requirements.......................................................29
3.4.1 Performance Requirements...........................................29
3.4.2 Safety Requirements.....................................................30
3.4.3 Security Requirements..................................................30
3.4.4 Software Quality Attribute............................................31
3.5 Analysis Model............................................................................32
3.6.1 Data Flow Diagrams.....................................................32
3.6.2 Class Diagrams.............................................................34
3.6.3 State-Transition Diagrams............................................34
3.7 System Implementation Plan.......................................................35

v
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

4 System Design...............................................................................................40 - 43
4.1System Architecture..................................................................40
4.2 Cicuit Diagram.........................................................................41
4.3 Block Diagram.........................................................................42
4.4UML Diagram...........................................................................43
5 Technical Specifications...............................................................................44 - 46
5.1 Technology details used in the project.....................................44
5.2 References to technology ........................................................46
6 Project estimate (Expected and Actual) Schedule (Sem I & Sem II ) and Team
structure(In the form of Diagram) .................................................................47 - 54
7 Software and Hardware Implementation....................................................55 - 65
7.1 Hardware Implementation........................................................55
7.2 Software Implementation.........................................................58
8 Software Testing............................................................................................66 - 70
8.1 Introduction...............................................................................66
8.2 Test cases (Unit test, Integration Test, Acceptance Test, Product Test)...66
8.3 snap shots of the test cases and test plans.................................70
9 Results and Discussion (Snap shots of the results) ....................................71 - 78
10 Deployment and Maintenance.....................................................................79 - 82
10.1 Deployment Strategies............................................................79
10.2 Maintenance Strategies...........................................................80
10.3 User Help................................................................................81
11 Conclusion and Future Scope.......................................................................83 - 85
References............................................................................................................86 - 87
Appendix .............................................................................................................88 - 99
Appendix A: Publications (if any) (In form of Index)
Published Paper in Hard Copy
Appendix B: Glossary [Define terms, acronyms, and abbreviations used]

vi
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

List of Figures

Name of Figures Pg No.

Figure 1 The Use case of Energy Harvesting and Data Collection 16
Figure 2 The Use case of Predictive Analysis and Real-Time Monitoring 17
Figure 3 The DFD level 0 32
Figure 4 The DFD level 1 33
Figure 5 The Class Diagram 34
Figure 6 The State Transition Diagram 34
Figure 7 The System Architecture 40
Figure 8 The Cicuit Diagram 41
Figure 9 The Block Diagram 42
Figure 10 The Activity Diagram 43
Figure 11 The Tile Construction 56
Figure 12 The cricuit plate 57
Figure 13 The Cercuit Plate 57
Figure 14 Full Hardware Setup 71
Figure 15 The Model Training 72
Figure 16 The Dashboard of Footstep data 73
Figure 17 Energy Forecasting 74
Figure 18 The Dashboard of Today vs Yesterday Comparison 74
Figure 19 The Monthly Predicted Data 75
Figure 20 The Discharged Energy to the Device 75
Figure 21 The AI vs Physics-Based Predicted Energy Graph 76
Figure 22 The Sensor Parameters Monitoring 77
Figure 23 The All UI 77

vii
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

List of Tables

Name Of Tables Pg No.

Table 1 A Litreture Survey Table 07
Table 2 Communication Components Table 28
Table 3 Requirement Gathering & Planning Table 47
Table 4 Hardware Setup & Data Collection Table 47
Table 5 Data Preprocessing & Feature Engineering Table 47
Table 6 AI Model Development Table 48
Table 7 Backend Integration Table 48
Table 8 Frontend Dashboard (Streamlit) Development Table 48
Table 9 Deployment and Hosting Table 49
Table 10 Testing and Validation Table 49
Table 11 Documentation and Final Report Table 49
Table 12 Risk Mitigation Plan Table 51
Table 13 Team Structure Table 51
Table 14 Task Hours Breakdown Table 52
Table 15 Cost Estimation Table 53
Table 16 Software Components Cost Table 54
Table 17 Summary of Estimated Project Cost Table 54
Table 18 Major Hardware Components Table 55
Table 19 Key Observations from Hardware Tests Table 58
Table 20 Data Dictionary 59
Table 21 Observation Table 70
Table 22 Troubleshooting Table 82

viii
 Chapter 1

Introduction
1.1Overview

The project focuses on developing an innovative Predictive Analysis Of Power Generation

From Footsteps Using Machine Learning that captures mechanical energy from human
footsteps and converts it into electrical power using piezoelectric sensors. This system targets
high-traffic areas such as railway stations, schools, malls, and public spaces, where
continuous footfall activity can generate sustainable energy. The harvested energy is suitable
for small-scale applications such as lighting LEDs or powering low-energy devices. The
proposed system integrates data collection, real-time processing, and energy storage modules
to create a seamless flow from energy capture to utilization. Predictive modeling techniques,
such as Linear Regression and Random Forest Regression, are employed to estimate energy
output based on footstep parameters, enhancing system efficiency and scalability.
A Python-based dashboard system further adds value by enabling real-time monitoring and
visualization of key metrics, such as pressure, frequency, and energy output. The project
leverages advanced tools and technologies, including Arduino, Python, and Pydash, for
system development. With a user-friendly interface and robust energy generation capabilities,
the system is designed to be scalable and eco-friendly. This project highlights the potential of
using untapped energy sources like human movement to contribute to the global energy grid,
offering a practical, low-cost solution to the energy crisis, especially in densely populated
urban areas.

1.2Brief Description

The project titled "Predictive Analysis of Power Generation from Footsteps Using Machine
Learning" presents an innovative approach to sustainable energy generation by harvesting
mechanical energy from human footsteps. Utilizing piezoelectric sensors, the system captures
pressure and vibrations caused by footfall in high-traffic areas such as railway stations,
shopping malls, schools, and other public spaces. This mechanical energy is then converted

1
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

into electrical energy, which can be stored and used for small-scale applications like lighting
LEDs or powering low-energy electronic devices.

A key feature of this system is its integration of machine learning models—specifically

Linear Regression and Random Forest Regression—to perform predictive analysis of energy
output. By analyzing parameters such as pressure, frequency of footsteps, and duration, the
models are trained to estimate the potential energy generation in real-time. This predictive
capability helps optimize system efficiency and makes the solution scalable for different
deployment environments.

The hardware component is built using Arduino microcontrollers, interfaced with

piezoelectric sensors for data acquisition. On the software side, a Python-based dashboard,
developed using tools such as PyDash and matplotlib, enables real-time visualization and
monitoring of critical metrics like pressure intensity, frequency of steps, and total energy
produced. This user-friendly interface not only aids in system maintenance but also provides
valuable insights for performance analysis and decision-making.

By merging sustainable energy harvesting with modern machine learning techniques, this
project offers a practical and eco-friendly solution to address energy demands in densely
populated urban areas. Its cost-effectiveness, low environmental impact, and scalability make
it an attractive option for smart city initiatives and energy-conscious infrastructures. Overall,
the system underscores the potential of transforming everyday human activities into useful
energy, contributing meaningfully to the global push for renewable and alternative energy
sources.

1.3Problem Definition

The increasing global energy demand, coupled with the depletion of conventional energy
sources, has necessitated the development of innovative, renewable energy solutions. In urban
environments, a significant amount of mechanical energy generated from human movement
remains untapped, contributing to energy wastage. Harnessing this energy effectively can
provide a sustainable and eco-friendly solution to meet small-scale power requirements,
especially in high-footfall areas like railway stations, malls, and public spaces. This project

2
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

seeks to address this problem by developing a footstep energy harvesting system that converts
human mechanical energy into electrical power while incorporating real-time monitoring and
predictive modeling for optimal performance.

1.4Objective

1. Develop a system that captures mechanical energy from footsteps using piezoelectric
sensors.
2. Efficiently convert the mechanical energy into electrical power for small-scale
applications like lighting or charging.
3. Implement real-time data monitoring of footstep pressure, frequency, and energy
output using a Python-based dashboard.
4. Use machine learning techniques such as Linear Regression and Random Forest to
predict energy output and optimize system performance.
5. Ensure the system is scalable, eco-friendly, and capable of future upgrades for broader
applications.
6. Identify and target locations with high foot traffic for maximum energy generation
efficiency.

1.5 Organization of Report

Chapter 1: Chapter 1 describes the project introduction of Predictive Analysis Of

Power Generation From Footsteps Using Machine Learning, which covers an overview,
brief description, problem definition, and objectives of the project in detail.

Chapter 2: Chapter 2 describes the research about the project and past development in
this project, and recent research about Predictive Analysis Of Power Generation From
Footsteps Using Machine Learning. It gives the research methodology & technology
used for this project in recent years.

Chapter 3: Chapter 3 describes system requirement elicitation, which is used to design

and develop Predictive Analysis Of Power Generation From Footsteps Using Machine
Learning. It covers all the functional & non-functional and design & development
requirements for this project.

3
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 4: Chapter 4 describes about system design, system architecture and UML
diagrams to design and develop Predictive Analysis Of Power Generation From
Footsteps Using Machine Learning.

Chapter 5: Chapter 5 describes about technical specification; tools & technologies we

are used to design and developed the system throughout the development life cycle. It
covers all the necessary tools we are used to developed Predictive Analysis Of Power
Generation From Footsteps Using Machine Learning.

Chapter 6: Chapter 6 describes the project plan that includes a detailed project schedule,
development team structure, and project estimation with cost estimation and resources
that we will utilize.

Chapter 7: Chapter 7 describes about the how we implemented the system and which are
the technologies and algorithms, we used to develop this system throughout the
development life cycle.

Chapter 8: Chapter 8 describes about the system testing and different tests that are
carried out on the system to cross verify the system functionality & performance of the
system.

Chapter 9: Chapter 9 describes the system development results with all the necessary
features of the system.

Chapter 10: Chapter 10 describes the system deployment and maintenance after the
development is completed.

4
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 2

Literature Survey

1. "Deep Learning Applications for Piezoelectric Power Systems"

This study emphasizes the use of deep learning, specifically Convolutional Neural Networks
(CNNs), for accurate predictions of energy generation in piezoelectric systems. The authors
showcase CNNs' ability to process complex step patterns, particularly in variable traffic
conditions, to enhance energy prediction accuracy, making these systems more adaptive and
efficient [1].

2. "Machine Learning Models for Predictive Power Generation in Piezoelectric

Systems"

This article applies decision trees and ensemble methods to forecast energy generation in
piezoelectric systems, focusing on high-traffic areas. The authors analyze variables like step
force and frequency, demonstrating how predictive models can improve energy harvesting
efficiency in environments such as transport stations and sports arenas [2].

3. "Renewable Energy Prediction Using Machine Learning: Trends and Challenges"

This paper discusses the application of machine learning techniques, including Naïve Bayes,
for renewable energy systems. Although focused broadly on renewable energy, the insights
are applicable to footstep-based piezoelectric systems, offering strategies for energy
prediction and optimization using traffic patterns and environmental data. The authors
highlight the role of predictive modeling in improving energy system performance [3].

4. "IoT-Integrated Piezoelectric Tiles for Real-Time Energy Prediction and Monitoring"

This study integrates IoT devices with piezoelectric tiles to monitor and predict energy
generation in real-time. Targeted at high-traffic areas such as shopping malls and hospitals,
the authors demonstrate how neural networks can predict power outputs using real-time data

5
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

from piezoelectric sensors. The integration of IoT and machine learning is shown to enhance
the continuous and efficient operation of energy harvesting systems [4].

5. "Energy Optimization Using Footstep Energy Harvesting Systems with Machine

Learning Models"

This research investigates the use of piezoelectric sensors for energy harvesting in public
spaces with significant foot traffic, such as train stations and airports. Using regression
models, the authors predict energy generation based on parameters like pressure and step
frequency. Their findings underline the role of machine learning in enhancing the efficiency
of these systems, offering insights into sustainable energy solutions for public infrastructure
[5].

6. "Piezoelectric Energy Harvesting for Smart Cities: A Machine Learning Approach"

This article explores harvesting energy from footsteps in urban environments using
piezoelectric sensors and machine learning techniques. The study highlights the ability of
piezoelectric tiles to convert mechanical pressure from footsteps into electrical energy,
emphasizing their potential for high-foot-traffic areas like public transportation hubs and
shopping malls. By applying machine learning algorithms, the authors predict and optimize
energy outputs based on variables such as step frequency and force, demonstrating the utility
of such systems for sustainable urban infrastructure [6].

6
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Author(s Year Study Page Relevance to AI Differences from Links

) Title Rang Your Project Technique Your Project
e

Lee and 2024 Deep Pg. Investigates Convolution Uses advanced Link
Park Learning 43-50 deep learning al Neural deep learning
Applicati methods for Networks techniques which
ons for enhancing (CNNs) may be
Piezoelect accuracy in used for computationally
ric Power energy recognizing intensive
Systems prediction, complex compared to your
relevant to patterns in real-time
your predictive step data. Arduino-based
model. system.

Sharma 2023 Machine Pg. Describes Decision Focused on Link

and Learning 134- predictive trees and predefined
Gupta Models 142 models for ensemble environments
for piezoelectric methods rather than
Predictive energy applied for dynamic inputs
Power generation, accurate such as weight
Generatio relevant to power and steps in real-
n in your goal of prediction. time.
Piezoelect predicting
ric power
Systems generation.

7
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Gil-Vera 2023 Renewabl Pg. Discusses ML- Applied Focuses on Link

and e Energy 72-80 based Naïve renewable energy
Quintero Prediction prediction of Bayes for sources like solar;
Using energy forecasting your project
Machine generation, trends, focuses
Learning: applicable to which can specifically on
Trends your power be applied piezoelectric
and prediction to energy
Challenge based on user predicting generation.
s input. power
generation.

Kumar et 2022 IoT- Pg. Explores IoT Neural IoT and network- Link
al. Integrated 89-97 integration for networks based, while your
Piezoelect real-time used for project uses
ric Tiles monitoring predicting Arduino without
for Real- and energy energy requiring a
Time prediction, outputs network-based
Energy akin to your based on system.
Prediction Arduino-based real-time
and real-time step data.
Monitorin system.
g

Shukla 2021 Energy Pg. Analyzes Regression Focuses on static Link

and Optimizat 121- footstep models for datasets rather
Yadav ion Using 130 energy predicting than real-time
Footstep harvesting power feedback and
Energy with generation visualization.
Harvestin optimization based on
g Systems using machine footstep
with learning data.
Machine models,
Learning applicable to

8
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Models your predictive

analysis.

Hossain 2020 Piezoelect Pg. Focuses on Applied Focused more on Link

et al. ric 45-52 energy Random large-scale urban
Energy harvesting Forest for deployment rather
Harvestin from footsteps predicting than personalized
g for using power systems or AI
Smart piezoelectric generation integration for
Cities: A sensors, based on dynamic learning.
Machine similar to your step
Learning real-time data frequency
Approach collection. and force.

Table 1 Litreture Survey Table

9
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 3

Software Requirements Specification

3.1 Introduction

The system requires both hardware and software components for effective energy harvesting
and predictive analysis. Hardware includes piezoelectric sensors and Arduino for capturing
footstep energy, while software uses Python and machine learning libraries for data
processing, prediction, and real-time dashboard visualization.

3.1.1 Purpose

The purpose of this system is to design and implement a smart energy generation solution that
captures mechanical energy from human footsteps using piezoelectric sensors and converts it
into usable electrical energy. The system further utilizes machine learning algorithms to
predict energy output based on footstep parameters, enabling real-time monitoring and
efficient energy utilization. This project aims to promote sustainable energy practices in high-
footfall areas such as railway stations, malls, and schools.

3.1.2 Project Scope

1. Energy Harvesting:
- Capture mechanical energy from human footsteps using piezoelectric sensors embedded in
floors.
- Convert the captured energy into electrical power for small-scale applications like lighting
LEDs and charging low-power devices.

2. Real-Time Monitoring:
- Provide real-time visualization of energy output, footstep pressure, and other key metrics
using a Python-based dashboard.

3. Predictive Modeling:

10
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

- Use machine learning models (Linear Regression and Random Forest Regression) to
predict energy output based on footstep data, enabling performance optimization.

4. Application Areas:
- High-footfall locations such as railway stations, schools, shopping malls, and public
spaces.

5. Scalability:
- Design the system to be scalable and modular, allowing for future expansion and
deployment in larger areas.

6. Eco-Friendly Solution:
- Promote sustainability by utilizing renewable energy sources and reducing dependency on
conventional resources.

3.1.3Design and Implementation Constrains

1. Hardware Constraints
Sensor Limitations: The piezoelectric sensors used to capture energy from footsteps have
specific sensitivity and power generation limits. The efficiency of these sensors may vary
depending on footstep intensity, frequency, and the type of surface.

Power Storage: Storing the harvested energy efficiently is a major constraint. Battery life,
charging cycles, and storage capacity must be considered to ensure that the system can hold
enough power to be useful for small-scale applications like LED lighting.

Microcontroller Limitations: The processing capability of microcontrollers, such as

Arduino, may limit the real-time processing of footstep data and energy prediction algorithms.
They may also have constraints in terms of memory and processing speed for handling
machine learning models.

2. Software Constraints

11
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Real-time Processing: The system is designed to handle real-time data, which can be
challenging due to the computational demands of processing footstep data and running
machine learning models (e.g., Linear Regression, Random Forest Regression) in real-time.

Data Accuracy: Machine learning models depend on accurate and clean data for prediction.
Noise in the sensor data due to environmental factors or sensor inaccuracies can reduce the
accuracy of energy predictions.

Software Integration: Integrating different software components like Python-based

dashboards, machine learning algorithms, and hardware interfaces requires careful
synchronization and compatibility. The system must ensure smooth communication between
sensors, microcontrollers, databases, and user interfaces.

3. Environmental Constraints
Footstep Frequency: The amount of energy generated depends heavily on foot traffic. In
locations with low foot traffic (e.g., at night or in less-populated areas), the system may not
generate enough power to be effective.

Weather Conditions: Environmental factors such as humidity, temperature, and dust can
affect the performance of piezoelectric sensors, potentially leading to degradation over time or
inaccurate energy measurements.

4. Budget and Resource Constraints

Cost: The overall cost of deploying piezoelectric sensors, microcontrollers, energy storage
units, and software development may be constrained by budget limitations. These constraints
could limit the number of sensors that can be deployed in large public areas.

Development Time: The time available for system development is limited, which means
certain features or optimizations may need to be postponed or simplified for initial
implementation. Machine learning model training and fine-tuning may also be time-
consuming and require sufficient computational resources.

5. Scalability Constraints

12
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

System Expansion: Scaling the system to handle large public areas with hundreds of sensors
can be challenging. Each sensor must be integrated into the central system, and the real-time
data processing and prediction must remain effective as the number of sensors increases.

Energy Harvesting Efficiency: As the system is scaled, it may be necessary to improve the
energy harvesting efficiency to meet growing demands. This could involve upgrading
piezoelectric sensor technology or finding better ways to store the generated energy.

6. User Interface Constraints

Accessibility: The user interface for monitoring and controlling the system must be easy to
use and accessible to non-technical users. However, simplifying the user interface could limit
the amount of detailed data or advanced functionalities available to more experienced users.

Real-time Visualization: The dashboard system should display real-time data, such as energy
output and foot traffic patterns, but the efficiency of the visualization may be limited by
processing power or bandwidth, especially in large-scale deployments.

3.1.4Assumptions and Dependencies

1. Assumptions

 Adequate Foot Traffic: It is assumed that the system will be deployed in areas with
consistent and adequate foot traffic, such as railway stations, malls, or schools. The
energy generated depends heavily on the frequency of footsteps, so locations with
minimal footfall may not generate enough power for practical use.
 Proper Sensor Placement: The system assumes that the piezoelectric sensors will be
properly installed in locations where they can effectively capture energy from
footsteps. The sensors need to be placed in areas with consistent foot traffic to ensure
reliable energy generation.
 Operational Environmental Conditions: It is assumed that the system will be
deployed in environments with relatively stable weather and temperature conditions.
Extreme conditions, such as high humidity, rain, or dust, may negatively affect sensor
performance, and additional protective measures may be needed.

13
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Stable Power Storage: The assumption is made that the energy storage devices (e.g.,
batteries or capacitors) used in the system will have sufficient capacity and lifespan to
store the energy generated during peak foot traffic periods and release it during off-
peak periods. The storage system is assumed to be able to handle the energy
fluctuations from sensor input.
 Reliable Data Collection: The system assumes that the piezoelectric sensors will
function reliably without significant interference or failure over time. It assumes that
sensors will be able to collect consistent and accurate data that can be processed to
predict energy output.
 Machine Learning Model Accuracy: The predictive models used (e.g., Linear
Regression, Random Forest Regression) are assumed to be able to generate accurate
predictions based on footstep parameters. The models are expected to be trained on
sufficient data to ensure their reliability in real-world scenarios.
 User Acceptance: It is assumed that the end-users (e.g., facility managers,
government bodies, or organizations) will accept and implement the system as a
sustainable energy solution in their premises.

2. Dependencies

 Hardware Components: The system's performance depends on the availability and

functionality of hardware components such as piezoelectric sensors, microcontrollers
(e.g., Arduino), and energy storage systems (e.g., batteries). The sensors must be
sensitive enough to detect the pressure exerted by footsteps, while the microcontrollers
must handle real-time data collection and processing.
 Software Libraries and Frameworks: The development of the machine learning
models relies on external software libraries, such as Python libraries like scikit-learn
for regression analysis and pandas for data manipulation. The project also depends on
libraries like matplotlib for data visualization and Pydash for real-time dashboard
creation.
 Cloud Services (Optional): If the system is integrated with cloud services for data
storage or remote monitoring, it will depend on the availability and reliability of these
cloud platforms. Additionally, if machine learning model updates are performed in the
cloud, this would create an additional dependency on cloud computing resources.

14
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Power Supply Reliability: The system assumes that there will be a stable power
supply to run microcontrollers and sensors. In some deployments, a backup power
system (such as a solar panel or UPS) may be required to ensure system continuity
during power outages.
 Internet Connectivity (For Real-time Monitoring): The system’s real-time
monitoring feature, including data visualization on the Python-based dashboard, may
depend on reliable internet connectivity, especially if remote access or cloud-based
storage is involved. Poor or unreliable internet service could limit system functionality.
 System Integration: The system depends on proper integration between various
components, such as sensors, microcontrollers, the energy storage system, and the
dashboard. Ensuring seamless communication and data exchange between these
components is crucial for system operation.
 Regulatory Compliance: The system must comply with local regulations regarding
energy harvesting, environmental safety, and data privacy. This includes legal
constraints on the installation and operation of energy-generating systems in public
spaces, and how the collected data is stored and processed.
 External Collaboration: The successful implementation of the system might depend
on partnerships with local authorities, businesses, or public institutions for installation,
maintenance, and system monitoring.

3.2 System Features (Use Case Diagrams)

3.2.1 System Feature 1: Energy Harvesting and Data Collection (Functional

Requirement)

Description:

The system must collect mechanical energy from human footsteps using piezoelectric sensors
and convert it into electrical energy. Simultaneously, it must collect essential data parameters
(such as pressure, force, footstep frequency) for further analysis.

15
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Functional Requirements:

 Detect each footstep using piezoelectric sensors.

 Measure pressure and force applied during each footstep.
 Convert mechanical energy into electrical energy.
 Store the collected energy into the energy storage unit (battery/capacitor).
 Record footstep-related data for machine learning model training and predictions.

Use Case Diagram:

Figure 1 The Use case Diagram of Energy Harvesting and Data Collection

Use Case Interaction:

 The user (pedestrian) steps on the sensor mat.

 The sensor detects the footstep, measures pressure/force.
 The system captures mechanical energy and converts it into electrical energy.

16
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Simultaneously, the system records the footstep parameters (pressure, time, frequency)
into a database for predictive modeling.

3.2.2 System Feature 2: Predictive Analysis and Real-Time Monitoring (Functional

Requirement)

Description:

The system must predict the amount of energy that will be generated based on live footstep
data using machine learning models (Linear Regression, Random Forest Regression). It must
also provide real-time monitoring and visualization through a user-friendly Python-based
dashboard.

Functional Requirements:

 Apply trained machine learning models on live footstep data.

 Predict energy generation based on footstep intensity, frequency, and pressure.
 Visualize real-time data (pressure applied, energy generated) on a dashboard.

Use Case Diagram:

Figure 2 The Use case Diagram of Predictive Analysis and Real-Time Monitoring

17
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Use Case Interaction:

 The system administrator or authorized user accesses the dashboard.

 The dashboard displays real-time graphs of footstep pressure, frequency, and energy
output.
 Machine learning models automatically predict the expected energy generation.
 The user can monitor the battery status and receive notifications if there is an
abnormality (e.g., no data from sensors or low battery levels).

3.3 External Interface Requirements

3.3.1User Interfaces

1. Inputs and Quick Statistics

The Sidebar Panel is a critical part of the system's user interface, offering users interactive
controls and immediate visibility into key daily energy statistics. It is designed to allow quick
adjustments and give an overview of system performance without navigating through multiple
screens.

1.1 Sensor Parameter Inputs

The sidebar provides sliders that allow users to adjust two major hardware-related parameters
to simulate different sensor or system conditions:

 Conversion Efficiency (%):

o Adjustable between 10% and 100% in steps of 5%.
o Represents the efficiency of converting the raw piezoelectric voltage into
usable electrical energy.
o Higher efficiency percentages simulate better energy harvesting technologies.
 Load Resistance (Ohms):
o Adjustable between 100 Ω and 10,000 Ω in steps of 100 Ω.
o Represents the resistance value across which the voltage is measured and the
power is dissipated.
o Varying the load resistance affects the calculated power and energy output.

18
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

1.2 Real-Time Statistics

In addition to parameter inputs, the sidebar displays real-time summary statistics based on
the current day's data file:

 Maximum Voltage Recorded (V):

o Displays the highest voltage generated from footstep events.
 Average Voltage (V):
o Calculates and displays the mean voltage recorded across all detected steps.
 Total Usable Energy (Physics Calculation) (Joules):
o Shows the amount of usable energy generated based on classical physics
equations, considering load resistance and voltage readings.
 Total Usable Energy (AI Prediction) (Joules):
o Displays the predicted usable energy, as calculated by a trained machine
learning model using voltage features.

These quick metrics provide an instant snapshot of system performance for the day without
requiring users to dive into detailed analysis.

2. Main Display Panel (Data Analysis and Visualization)

The Main Display Panel is the primary workspace for users. It presents detailed information,
analytics, visualizations, forecasting results, and energy usage suggestions.

2.1 Live Data File Information

 The system identifies and displays the name of the current data file (typically
representing today’s date).
 This ensures users know exactly which day’s footstep energy data they are viewing
and analyzing.

2.2 Today's Energy Analysis

This section summarizes the energy harvesting performance for the current day:

 Physics-Based Energy Calculation:

o Displays the cumulative energy harvested calculated using the formula:

19
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Energy(J)=Voltage2Resistance×ΔtEnergy (J) = \frac{Voltage^2}{Resistance}

\times \Delta tEnergy(J)=ResistanceVoltage2×Δt

o Adjusted based on the selected conversion efficiency.

 AI Model Predicted Energy:
o Displays the total usable energy predicted by a trained AI model, based on
voltage characteristics such as moving averages and rates of change.

2.3 Energy Forecasting

The system estimates the future energy generation potential:

 Average Energy per Step:

o Calculates the average amount of energy generated for each detected footstep
event.
 Total Steps Detected:
o Counts the number of footstep events detected by analyzing voltage spikes.
 Predicted Monthly Energy Generation:
o Projects the total energy that could be generated over a 30-day period,
assuming similar daily performance.

2.4 Energy Usage Recommendation

The dashboard provides suggestions on how the generated energy can be used:

 Usage Options:
o Powering an LED Light: Calculates how long an LED light could remain on
based on the available energy.
o Running a DC Motor Fan: Estimates the runtime of a small fan.
o Charging a Rechargeable Battery: Simulates the charging process of a 3V
battery and displays the charging percentage.
 Outputs:
o Provides the estimated runtime (in minutes) or battery charge percentage.
o Displays progress bars and informative messages for better visualization.

20
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

2.5 Real-Time Visualization

This section provides dynamic charts for visual analysis of system performance:

 Voltage over Time:

Line chart showing the variation of voltage throughout the day.
 Physics Energy Accumulation:
Cumulative graph of energy calculated using classical physics methods.
 AI-Predicted Energy Accumulation:
Cumulative graph of the AI-predicted energy over time.
 Instantaneous Power Output:
Real-time line chart showing the power (in Watts) generated at each recorded time
step.

These visualizations help users detect trends, patterns, and anomalies easily.

2.6 Multi-Day Comparison

The dashboard allows side-by-side comparisons between today's and yesterday's data:

 Users can select a metric to compare:

o Voltage values.
o Physics-based cumulative energy.
o AI-predicted cumulative energy.
o Instantaneous power output.
 Comparison Visualization:
o Dual-line charts are generated, one line representing today's data and the other
representing yesterday's.
o Helps in assessing day-to-day performance improvements or deviations.

2.7 Download Processed Data

 After processing and analyzing the data, the system allows users to download the
output.
 The processed data from today and yesterday are saved into an Excel workbook with
separate sheets.

21
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 This feature is useful for record-keeping, detailed offline analysis, and academic
reporting.

3. User Interaction Flow

The complete interaction between the user and the system follows a clear and logical
sequence:

1. Footstep on Piezoelectric Sensor:

o Users step on the specially designed piezoelectric tile or setup.
o The sensor converts the mechanical energy into electrical voltage.
2. Arduino Data Recording:
o An Arduino microcontroller continuously reads voltage values from the sensor.
o The data is timestamped and saved into a CSV file at regular intervals.
3. Dashboard Data Loading:
o The Streamlit dashboard automatically identifies the latest available CSV file.
o It loads the data for further processing.
4. Data Processing:
o Basic noise filtering is applied (removal of voltages below 0.01V).
o Physics-based and AI-based energy calculations are performed.
5. Parameter Adjustments:
o The user can modify conversion efficiency and load resistance values through
sidebar sliders.
o The system instantly updates the calculations and visualizations.
6. Data Visualization and Analysis:
o The dashboard displays real-time graphs, energy calculations, and performance
forecasting.
7. Energy Usage Simulation:
o The user selects how to use the available energy (light, fan, battery).
o The dashboard estimates device runtime or battery charge.
8. Multi-Day Performance Comparison:
o The system allows comparing today’s energy generation performance against
yesterday’s.
9. Data Export:

22
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o The user can download the processed data as an Excel file for future reference
or extended analysis.

3.3.2 Hardware Interfaces

1. Hardware Components

The main hardware elements used in the system are:

1.2 Piezoelectric Sensor (Footstep Tile)

 Function:
Converts mechanical pressure from footsteps into electrical voltage.
 Working Principle:
When mechanical stress (from a footstep) is applied to the piezoelectric material, it
generates an electrical charge proportional to the force.
 Output:
Raw AC voltage signal which needs rectification for stable reading.

1.2 Rectifier Circuit (AC to DC Converter)

 Function:
Converts the AC output of the piezoelectric sensor into a stable DC voltage.
 Components:
o Full-Bridge Rectifier (e.g., diode bridge)
o Capacitor (for smoothing voltage)
 Purpose:
Ensures that the Arduino reads a clean, stable DC voltage signal.

1.3 Arduino Uno (Microcontroller)

 Function:
Acts as the central hardware controller.
 Roles:
o Reads the analog voltage values from the sensor.
o Time-stamps each reading.

23
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o Sends the collected data to a computer via serial communication (USB).

o Optionally saves data into a connected SD card if standalone operation is
required.

1.4 Computer or Laptop

 Function:
Receives voltage data sent by Arduino over the USB connection.
 Purpose:
Runs Python scripts or Streamlit Dashboard to read CSV data files, analyze energy
generation, and visualize results.

1.5 Optional Components:

 LCD Display (I2C module):

o Displays real-time voltage or energy values during operation.
 Rechargeable Battery and Power Bank Circuit:
o Stores harvested energy for later use.

2. Hardware Setup and Data Flow

1. Footstep Impact:
o User steps onto the piezoelectric tile, generating a small voltage pulse.
2. Rectification:
o The raw AC voltage from the piezoelectric sensor is passed through a rectifier
circuit to produce stable DC voltage.
3. Voltage Measurement:
o The rectified DC voltage is optionally passed through a voltage divider and fed
into the Arduino’s analog input pin (e.g., A0).
4. Arduino Processing:
o Arduino reads the analog voltage signal at a fixed sampling rate (e.g., every
100 milliseconds).
o Each voltage value is recorded along with a timestamp.
5. Data Transfer:

24
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o Arduino sends the voltage-time data to the computer via the serial USB
connection.
6. Data Storage:
o A Python script running on the computer listens to the serial port.
o The script saves the incoming data automatically into a CSV file, named with
the current date (e.g., 27-04-2025.csv).
7. Data Analysis and Visualization:
o The Streamlit dashboard reads the latest CSV file.
o It performs energy calculations, AI predictions, forecasting, visualization, and
provides energy usage recommendations.

3.3.3 Software Interfaces

1.Software Components

The key software components of the system include:

1.2. Arduino IDE

 Purpose:
To program the Arduino Uno to read analog voltage values and send the data through
the serial port.
 Functions:
o Reads voltage from analog pin (A0).
o Time-stamps each reading.
o Sends the readings in a serial format (timestamp, voltage) to the connected
computer.

1.2. Python (Data Acquisition Script)

 Purpose:
To listen to the Arduino's serial output and save the incoming data.
 Libraries Used:
o pyserial — to read data from the serial port.
o pandas — to structure and store the data in tabular form.

25
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o datetime — to generate file names based on the current date.

 Functions:
o Receives live voltage-time data from Arduino.
o Automatically creates and saves a new CSV file daily (e.g., 27-04-2025.csv).

1.3. Machine Learning Model (Random Forest Regressor)

 Purpose:
To predict usable energy generation more accurately based on real-world sensor data.
 Libraries Used:
o scikit-learn — for training and deploying the Random Forest model.
o joblib — for saving and loading the trained model.
 Model Inputs (Features):
o Instant voltage.
o Moving average of voltage.
o Voltage change between samples.
o Rate of voltage change relative to moving average.
 Model Output (Target):
o Predicted instantaneous energy in Joules.

1.4. Streamlit Dashboard (User Interface)

 Purpose:
To provide a real-time interactive graphical dashboard for analysis, forecasting, and
decision-making.
 Libraries Used:
o streamlit — for building the web app dashboard.
o pandas — for handling CSV and processed data.
o numpy — for calculations.
o xlsxwriter — for exporting data to Excel files.
 Functions:
o Loads today's and yesterday’s CSV data files.
o Applies physics-based and AI-based energy calculations.
o Provides forecasting estimates and energy usage suggestions.
o Displays real-time charts of voltage, power, and energy accumulation.

26
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o Allows downloading processed data for external use.

2. Software Workflow

1. Data Acquisition:
o Arduino reads analog voltage from the piezoelectric sensor.
o Arduino sends timestamped voltage data to the computer.
2. Data Storage:
o A Python script listens to the Arduino serial output.
o It saves incoming data into a CSV file automatically.
3. Data Loading:
o The Streamlit dashboard loads the most recent CSV file(s).
4. Data Processing:
o Cleans noise by removing very low voltages (<0.01V).
o Calculates real-time power and physics-based energy.
o Applies the trained machine learning model to predict energy.
5. Visualization and Interaction:
o Displays real-time graphs of voltage, energy, and power.
o Estimates average energy per step and total steps.
o Forecasts monthly energy generation.
o Suggests energy usage (e.g., powering a light, motor, or battery charging).
o Provides option to download processed data in Excel format.

3.3.4 Communication Interfaces

1.Communication Type

 Mode: Serial Communication (Wired)

 Protocol: UART (Universal Asynchronous Receiver-Transmitter) over USB

27
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

2.Communication Components

COMPONENT DESCRIPTION
Arduino Uno Captures analog voltage from piezoelectric sensor and sends serial
data via USB.
USB Cable Connects Arduino to the computer for power and data transfer.
Computer (PC/Laptop) Receives serial data from Arduino, processes it with Python, and
displays results on the Streamlit dashboard.
Serial Monitor / Reads, decodes, and saves the incoming serial data into structured
Python Script CSV files.

Table 2 Communication Components Table

3.Data Transmission Format

Each data packet sent by the Arduino follows a simple, consistent format:

 Timestamp:
o Elapsed time in seconds (or milliseconds) since Arduino started.
o Helps maintain the sequence of data points.
 Voltage:
o Analog voltage measured at that time step.
o Represents the energy generated by footsteps on the piezoelectric sensor.

Example of Serial Output:

plaintext
CopyEdit
0.00, 0.014
0.10, 0.030
0.20, 0.009
0.30, 0.025

28
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

4. Communication Flow

1. Sensor Activation:
o A footstep on the piezoelectric sensor generates a voltage pulse.
2. Voltage Reading:
o Arduino reads the voltage via an analog input (e.g., A0 pin).
3. Data Formatting:
o Arduino formats the data as timestamp, voltage.
4. Data Transmission:
o Arduino sends the formatted data through its serial port via USB.
5. Data Reception:
o The computer’s Python script continuously reads incoming serial data.
6. Data Storage:
o Each reading is parsed and saved into a CSV file for further analysis.
7. Data Usage:
o The Streamlit dashboard loads the CSV file to perform analysis, forecasting,
and visualization.

3.4 Nonfunctional Requirements

3.4.1 Performance Requirements

The system must process real-time footstep data and update the dashboard within 2
seconds of input.

Predictive model results (Random Forest Regression) must be generated in under 5

seconds after receiving footstep parameters.

The energy conversion efficiency from mechanical to electrical energy should be greater
than 70% based on optimal footstep force.

Data recording (pressure, frequency, time) should happen without data loss even during
continuous high footfall rates (~100 steps/min).

29
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

The dashboard should support at least 10,000 recorded footsteps without performance
degradation.

3.4.2 Safety Requirements

Piezoelectric tiles must be shock-proof and capable of handling human weight up to 100
kg without damage.

All wiring and exposed electronics must be properly insulated to prevent electric shock
hazards.

The generated voltage levels must be kept within safe limits (below 12V) for human
interaction.

Backup power should ensure that if the main system fails, sensor data is safely stored
without corruption.

The physical platform must have anti-slip surfaces to prevent user injury.

3.4.3 Security Requirements

Only authorized users (administrators) must be allowed to access the dashboard settings
and system configurations.

Data transmitted between Arduino and the Python dashboard (if over a network) should
use secured communication protocols (e.g., encrypted serial or wireless data if applicable).

User authentication (e.g., simple password protection) should be implemented for system
maintenance modes.

Regular backup of collected sensor and prediction data to prevent data loss during system
failure.

Protection mechanisms against tampering with hardware components (sealed wiring,

tamper-evident seals).

30
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

3.4.4 Software Quality Attribute

Reliability:
The system must operate continuously with minimal downtime during heavy footfall events.

Scalability:
The system design should allow scaling by adding more piezoelectric tiles or connecting
multiple sensor mats.

Usability:
The dashboard interface should be intuitive and easy to navigate for monitoring energy
generation without specialized training.

Maintainability:
Codebase (Python and Arduino) should be modular, well-commented, and documented for
easy updates or bug fixing.

Portability:
The software solution (dashboard and predictive model) should be deployable on different
machines (Windows/Linux) with minimal setup.

Efficiency:
The system should consume minimal computational and energy resources while running
predictive models and visualizations.

31
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

3.5 Analysis Model

3.5.1Data Flow Diagrams

1. DFD Level 0 Diagram:

Figure 3 The DFD level 0

32
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

2. DFD Level 1 Diagram:

Figure 4 The DFD level 1

33
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

3.5.2 Class Diagrams

Figure 5 The Class Diagram

3.5.3State-Transition Diagrams

Figure 6 The State Transition Diagram

34
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

3.6 System Implementation Plan

3.6.1 Data Collection Module:

The Data Collection Module is responsible for collecting raw sensor data required for
building and training the AI model for energy prediction.

This module includes:

 Setting up piezoelectric sensors to collect voltage and step count data.

 Capturing real-time voltage signals generated through mechanical pressure (steps).
 Logging the collected data securely for preprocessing and modeling.

Hardware Setup:

 Use an Arduino board to interface with piezoelectric sensors.

 Attach sensors to a surface where mechanical pressure (steps) is applied.
 Calibrate sensors for accurate data collection.
 Ensure stable power supply and data logging for smooth operation.

Key Tasks:

 Hardware setup and calibration (Piezo sensor + Arduino).

 Real-time data acquisition through Arduino.
 Data logging in CSV/Database.

3.6.2 Data Preprocessing Module:

The Data Preprocessing Module ensures that the collected data is clean, consistent, and
ready for machine learning model training.

This module includes:

 Handling missing values and noise in the collected sensor data.

 Smoothing signals (e.g., using Moving Average) and normalizing values.
 Feature extraction: calculating voltage change rate, moving average, step counts.

35
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Key Tasks:

 Data cleaning and noise reduction.

 Feature engineering (like Voltage_MA, Voltage_Change, etc.).
 Preparing the final dataset for AI model training.

3.6.3 Machine Learning Model Development Module:

The Machine Learning Model Development Module focuses on creating a predictive AI

model to estimate the energy generated from piezoelectric sensors.

This module includes:

 Selecting appropriate machine learning models (e.g., Random Forest Regressor,

XGBoost).
 Splitting data into training and testing sets.
 Training, tuning, and evaluating the models using RMSE, MAE, and R² metrics.

Key Tasks:

 Model selection (Random Forest, XGBoost).

 Model training and hyperparameter tuning.
 Model evaluation and validation.
 Saving the trained model (.pkl file).

3.6.4 Backend Development Module:

The Backend Development Module manages the communication between Arduino, AI

Model, and Dashboard.

This module includes:

 Developing a Python backend using Streamlit to receive sensor data.

 Loading the trained AI model for live predictions.
 Sending prediction results to the frontend dashboard.

36
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Key Tasks:

 Backend development using Streamlit.

 Integration of AI model with backend.
 Securing backend endpoints for safe data communication.

3.6.5 Frontend Development Module:

The Frontend Development Module provides a clean user interface for displaying real-time
energy generation results.

This module includes:

 Developing a Streamlit dashboard.

 Displaying real-time voltage readings, step counts, and predicted energy.
 Graphical visualization of sensor data trends and AI predictions.

Key Tasks:

 Streamlit dashboard UI/UX design.

 Graph plotting (Voltage vs Time, Energy vs Steps).
 Integration with backend APIs for live data updates.

3.6.6 Database Development Module:

The Database Development Module stores sensor data, model outputs, and user interaction
records.

This module includes:

 Selecting a lightweight database (e.g., SQLite or Firebase) to store incoming data.

 Designing the database schema to store voltage, steps, timestamp, and predicted
energy values.
 Managing database security and backups.

Key Tasks:

 Database design and setup.

37
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Real-time data insertion and retrieval.

 Implementing backup and security measures.

3.6.7 Testing & Debugging Module:

The Testing & Debugging Module ensures the robustness of the full system (hardware +
software).

This module includes:

 Unit testing of backend APIs and frontend dashboard.

 End-to-end testing from sensor data input to AI prediction display.
 Debugging hardware issues like false readings or data loss.

Key Tasks:

 Backend and frontend unit testing.

 Hardware and software integration testing.
 Debugging and optimization.

3.6.8 Deployment & Hosting Module:

The Deployment & Hosting Module makes the system production-ready and available for
users.

This module includes:

 Deploying the Streamlit dashboard on cloud platforms (e.g., AWS, Streamlit Cloud, or
Heroku).
 Setting up Arduino with continuous serial data streaming.
 Ensuring system scalability for real-world use.

Key Tasks:

 Deployment of backend and frontend on cloud.

 Setting up continuous integration and delivery (CI/CD).
 Monitoring deployed system for reliability.

38
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

3.6.8 Documentation & Reporting:

The Documentation & Reporting Module ensures complete documentation of the system
design, codebase, and user manual.

This module includes:

 Writing technical documentation of system modules.

 Preparing user guides for system operation.
 Final project report including challenges faced and how they were overcome.

Key Tasks:

 Technical documentation (code + hardware setup).

 User manual preparation.
 Final project report and presentation.

3.6.9 Maintenance & Technical Support:

The Maintenance & Technical Support Module provides continuous system upkeep and
user assistance.

This module includes:

 Regular system checks and updates (software and hardware).

 Addressing any user-reported issues post-deployment.
 Adding improvements based on user feedback.

Key Tasks:

 System monitoring and bug fixes.

 Providing helpdesk support.
 Implementing system updates and feature improvements.

39
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 4

System Design

4.1System Architecture

Figure 7 The System Architecture

40
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

4.2Cicuit Diagram

Figure 8 The Cicuit Diagram

41
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

4.3Block Diagram

Figure 9 The Block Diagram

42
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

4.4UML Diagram

1. Activity Diagram:

Footstep Detection

Convert Pressure to
Voltage

Process Data

Predict Power
Output

Display Output

Store Excess Energy

Use Power or
Indicate Status

Figure 10 The Activity Diagram

43
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 5

Technical Specifications

5.1 Technology details used in the project

5.1.1 Hardware Components:

1. Piezoelectric Sensors:
o Role: Converts mechanical pressure (e.g., footsteps) into electrical energy.
o Key Features: High sensitivity, ability to harvest energy from pressure events,
such as footfalls, for energy production.
2. Arduino:
o Role: Microcontroller used to interface with piezoelectric sensors, collect
sensor data, and transmit data to the backend.
o Key Features: Simple setup, flexibility for multiple sensors, supports Arduino
IDE for easy programming.
3. Voltage Measurement System:
o Role: Measures real-time voltage generated by the piezoelectric sensors,
providing the necessary input for further energy prediction.
o Key Features: Accurate voltage reading to enable real-time data analysis for
energy prediction.
4. Computer/Server (for AI Model and Backend):
o Role: Hosts the AI model and manages backend processes for data processing,
predictions, and communication with the front-end dashboard.
o Key Features: High computational power to handle machine learning tasks,
running Python and related machine learning libraries.

5.1.2 Software Components:

1. Python:
o Role: Main programming language for AI model development, data
preprocessing, and backend processing.

44
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o Key Libraries: pandas (data manipulation), numpy (numerical operations),

scikit-learn (machine learning models), xgboost (gradient boosting models).
2. Machine Learning Frameworks:
o Random Forest Regressor and XGBoost are used for predicting energy
generation based on sensor data.
o Key Features: These models are capable of handling regression tasks,
identifying patterns from input data, and providing accurate energy predictions.
3. Streamlit:
o Role: Used for creating the interactive web dashboard to display real-time
sensor data and AI predictions.
o Key Features: Simple, user-friendly interface for real-time updates,
visualizations, and integration with Python scripts.
4. CSV Files for Data Storage:
o Role: Stores the sensor data, energy predictions, and relevant timestamps in
CSV files.
o Key Features:
 Easy to implement: No complex database setup is required.
 Lightweight: Suitable for small to medium-scale data storage needs.
 Portable: Data can easily be opened and analyzed in common tools like
Excel or Python.
 Scalability Considerations: While efficient for smaller data volumes,
handling large datasets may require additional strategies, such as
splitting into multiple files or periodic backups.

Data Stored in CSV:

 Sensor Data: Includes voltage readings, step counts, and timestamps.

 Predicted Energy: The energy calculated by the AI model based on
the input sensor data.
5. Cloud/Hosting Platforms:
o Role: Deployment platform for the Streamlit dashboard.
o Platforms Used: AWS, Streamlit Cloud, or Heroku.
o Key Features: Scalable, reliable, and able to host interactive web applications.

45
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

5.2 References to technology

1. Piezoelectric Energy Harvesting:

o Reference: D. A. M. H. R. Y. S. (2019). "Energy Harvesting from Piezoelectric
Materials". Journal of Applied Energy. DOI: 10.1016/j.apenergy.2019.113504.
2. Arduino:
o Reference: Arduino. "Arduino Official Website". Available at:
https://www.arduino.cc/.
3. Python & Machine Learning Libraries:
o Reference: Python Software Foundation. "Python Programming Language".
Available at: https://www.python.org/.
o Reference: R. J. A. (2020). "Machine Learning with Scikit-Learn and
XGBoost". Journal of Machine Learning. DOI: 10.1109/TMLR.2020.2954236.
4. Streamlit:
o Reference: Streamlit Inc. "Streamlit Official Website". Available at:
https://streamlit.io/.
5. CSV for Data Storage:
o Reference: Python Software Foundation. "pandas Documentation: I/O Tools".
Available at: https://pandas.pydata.org/pandas-
docs/stable/reference/api/pandas.DataFrame.to_csv.html.
6. Cloud Deployment:
o Reference: Heroku. "Heroku Cloud Platform". Available at:
https://www.heroku.com/.
o Reference: AWS. "Amazon Web Services for Hosting". Available at:
https://aws.amazon.com/.

46
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 6

Project estimate Schedule and Team structure

6.1 Project Plan (Expected vs Actual)

Phase 1: Requirement Gathering & Planning

EXPECTED ACTUAL
Timeline 1 Week 1 Week
Activities Identify hardware and software Completed as planned. All
requirements (Piezo sensor, Arduino, necessary components were
Python, Streamlit, ML libraries). Plan data finalized, and a basic system
collection methods and model selection architecture was outlined.
strategies.
Challenges Minor risks expected in selecting the right No major challenges
sensor configuration. encountered. Clear understanding
of project scope early helped
smooth planning.
Table 3 Requirement Gathering & Planning Table

Phase 2: Hardware Setup & Data Collection

EXPECTED ACTUAL
Timeline 2 Weeks 2.5 Weeks
Activities Setup Arduino with piezoelectric Slight delay due to calibration issues
sensors. Collect voltage and of sensors. However, successful data
mechanical force data under controlled collection achieved with real-time
conditions. Store data in CSV files. logging into CSV.
Challenges Possible sensor calibration difficulties Faced slight inconsistencies in sensor
and unstable readings. readings initially; solved by
recalibrating Arduino code and
rechecking wiring.
Table 4 Hardware Setup & Data Collection Table

Phase 3: Data Preprocessing & Feature Engineering

EXPECTED ACTUAL
Timeline 1 Week 1 Week
Activities Clean data, remove noise, compute features Completed as planned. Developed
like Moving Average (Voltage_MA), a robust data preprocessing script
Voltage Change, and Voltage Change Rate. with pandas. Feature extraction
Prepare the final dataset for model training. was smoothly integrated.
Challenges Handling noisy signals. No major issues; Moving Average
smoothing handled noise

47
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

effectively.
Table 5 Data Preprocessing & Feature Engineering Table

Phase 4: AI Model Development

EXPECTED ACTUAL
Timeline 2 Weeks 2 Weeks
Activities Train machine learning models (Random Completed within expected time.
Forest, XGBoost). Split into train-test sets. Random Forest Regressor
Evaluate models with RMSE, MAE, R². performed best and was saved for
Save the best-performing model (.pkl). deployment.
Challenges Model overfitting concerns, selecting right Minor hyperparameter tuning was
features. needed to optimize performance.
No major delays.
Table 6 AI Model Development Table

Phase 5: Backend Integration

EXPECTED ACTUAL
Timeline 1 Week 0.5 Week
Activities Develop backend to read real-time Faster than expected because backend
sensor data, load the AI model, and logic was simple. No Flask; directly
prepare data for dashboard. integrated inside Streamlit app.
Challenges Possible API delays if using a server- Streamlit-only integration avoided
client model. complex backend API development,
saving time.
Table 7 Backend Integration Table

Phase 6: Frontend Dashboard (Streamlit) Development

EXPECTED ACTUAL
Timeline 1.5 Weeks 2 Weeks
Activities Create a Streamlit dashboard showing Slight delay due to UI optimization
live voltage, step counts, predicted and making graphs more dynamic.
energy, and graphs (Voltage vs Time, Final dashboard was responsive and
Energy vs Steps). user-friendly.
Challenges Real-time updates and graph Solved by using Streamlit's
responsiveness. st.line_chart() and caching
techniques to manage live updates.
Table 8 Frontend Dashboard (Streamlit) Development Table

48
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Phase 7: Deployment and Hosting

EXPECTED ACTUAL
Timeline 1 Week 1 Week
Activities Host the Streamlit app on a cloud platform Completed smoothly. Streamlit
like AWS, Streamlit Cloud, or Heroku. Cloud was selected for simplicity.
Ensure Arduino keeps sending data.
Challenges Hosting latency issues. No major issues. Minor
adjustments needed for Streamlit
Cloud settings (session timeouts).
Table 9 Deployment and Hosting Table

Phase 8: Testing and Validation

EXPECTED ACTUAL
Timeline 1 Week 1 Week
Activities End-to-end system testing (Hardware → Testing was successful. Minor
Streamlit Dashboard). Check for voltage bugs in data formatting were
reading accuracy, prediction correctness, resolved. Prediction accuracy
and graph updates. matched expectations.
Challenges Handling unexpected sensor Added try-except blocks in
disconnections. Arduino and Python scripts to
handle disconnections gracefully.
Table 10 Testing and Validation Table

Phase 9: Documentation and Final Report

EXPECTED ACTUAL
Timeline 1 Week 1 Week
Activities Prepare technical documentation, Completed as planned. Detailed
user manuals, project report, and documentation prepared including setup
PPT for presentation. guide, user manual, and performance
graphs.
Challenges Capturing hardware-software Solved by including block diagrams and
interactions clearly. sample data screenshots in the
documentation.
Table 11 Documentation and Final Report Table

49
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

6.2 Risk Mitigation Plan

RISK IMPACT MITIGATION STRATEGY

Sensor Calibration Inaccurate voltage Perform multiple calibration cycles.
Errors readings leading to poor Validate sensor outputs with multimeter.
AI model performance. Implement software-based filtering
(Moving Average).
Noisy or Incomplete Poor model training and Use noise reduction techniques during
Data unreliable predictions. preprocessing. Log data carefully and
discard corrupted samples.
Model Overfitting Low generalization on Apply cross-validation, hyperparameter
unseen data. tuning, and feature selection. Monitor
evaluation metrics (RMSE, MAE, R²).
Arduino System failure during Add error handling (try-except blocks)
Disconnection real-time prediction. in the code. Reconnect automatically if
the serial connection drops.
Streamlit Dashboard Poor user experience due Optimize data reading intervals, use
Lag to slow graph updates. Streamlit’s caching and efficient plotting
libraries.
Hosting/Deployment System downtime or Choose reliable cloud hosting (Streamlit
Issues inaccessibility. Cloud). Keep a local backup version
ready.
Delay in Hardware Delay in overall timeline. Start hardware setup early in the project
Assembly phase. Keep spare parts available for
quick replacement.
Table 12 Risk Mitigation Plan Table

6.3 Team Structure

TEAM MEMBER ROLE RESPONSIBILITIES

- Project Leader - Overall project management and
Hemant Chauthe: - Machine Learning coordination.
Team Lead & AI Model Specialist - Collecting and preparing datasets.
Developer - Building, training, tuning, and evaluating
the AI models (Random Forest, etc.).
- Saving the final model (.pkl) and
integrating it with the Streamlit app.
- Managing the timeline and ensuring
milestones are achieved.
- Hardware Engineer - Designing the hardware setup using
Dhananjay Patil: piezoelectric sensors and Arduino.
Hardware Specialist - Calibrating sensors for accurate voltage
measurements.
- Writing Arduino code for real-time data
collection.
- Sending live data to the Streamlit
application via serial communication.

50
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

- Handling sensor maintenance and

troubleshooting hardware issues.
- UI/UX and - Designing and developing a user-friendly
Harshada Pawar: Visualization Streamlit dashboard.
Frontend Developer Developer - Visualizing real-time voltage, step count,
and predicted energy data.
- Plotting graphs (Voltage vs Time, Energy
vs Steps) neatly.
- Improving UI/UX for smoother
performance.
- Integrating frontend components with
backend APIs (if any).
- Data Processor and - Writing preprocessing scripts (feature
Revati Jadhav: Quality Assurance extraction, moving averages).
Backend Integrator & - Handling loading of the AI model inside
Tester Streamlit (no separate Flask backend used).
- Testing end-to-end system from data
input → AI model → dashboard output.
- Debugging issues in data flow, AI
prediction, or Streamlit display.
- Conducting final testing and ensuring
system robustness before deployment.
Table 13 Team Structure Table

6.4 Task Hours Breakdown (Estimated)

TEAM MEMBER TASK ESTIMATED

HOURS
Hemant Chauthe (Team Lead Data Collection Supervision 6 hours
& AI Developer)
Data Preprocessing and Feature 8 hours
Engineering
Model Training and Tuning 10 hours
(Random Forest)
Model Saving and Integration (with 4 hours
Streamlit)
Project Coordination and Progress 4 hours
Review
Subtotal 32 hours
Dhananjay Patil (Hardware Arduino + Sensor Setup 8 hours
Specialist)
Sensor Calibration and Testing 6 hours
Arduino Code Writing and Data 6 hours
Logging (CSV)
Hardware Debugging and 4 hours
Maintenance
Subtotal 24 hours

51
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Harshada Pawar (Frontend Streamlit Dashboard UI/UX Design 8 hours

Developer)
Real-time Graph Development 6 hours
(Voltage-Time, etc.)
Dashboard Integration Testing 4 hours
Optimizing Dashboard Performance 2 hours
Subtotal 20 hours
Revati Jadhav (Backend Data Preprocessing Scripts 6 hours
Integrator & Tester)
Streamlit Model Loading & Live 6 hours
Prediction Handling
Full System Testing (End-to-End 8 hours
Testing)
Bug Fixing and Debugging 4 hours
Subtotal 24 hours
Table 14 Task Hours Breakdown Table

6.5 Cost Estimation (For 2 Tiles)

This project uses piezoelectric-based tiles integrated with a dashboard and predictive
analysis. The cost estimation for 2 tiles is as follows:

6.5.1 Hardware Components Cost

Sr. Component Specifications Quantity for Est. Cost Total

No. 2 Tiles per Unit Cost (₹)
(₹)

1 Piezoelectric 35mm, ceramic, 36 12 432

Transducer quartz crystal

2 Full-Wave Bridge W10 36 15 540

Rectifier

3 Diodes 1N4007, rectifying 36 2 72

diode

4 Capacitor 4700µF, 25V 2 50 100

5 2-Core Wire Red and black, 20 1 500 (for 500

meters full 60m

52
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

roll)

6 Wooden Tiles 30cm × 30cm 2 500 1,000

7 PCB Plate Copper clad 2 100 200

8 Nut-Bolt 8mm, mild steel 8 10 80

9 Battery Lead Acid, 12V, 1 1,000 1,000

1.3AH

10 LED 12V COB power 2 80 160

module

11 LCD 16x2 cm, 5V 1 250 250

12 Arduino UNO ATmega328, 1 800 800

16MHz

13 Jumper Wires Male-male, male- 15 10 150

female, 20cm

14 Breadboard 7-47 inch 1 150 150

15 Resistors 10KΩ, 100KΩ 2 5 10

16 Variable Resistor 10KΩ 1 20 20

17 DC Voltmeter 0–30V 1 300 300

Table 15 Cost Estimation Table

Total Hardware Cost ≈ ₹5,764

6.5.2 Software Components Cost

Item Description Estimated Cost (₹)

Python (Anaconda) Open-source tool used for coding, 0 (Free)
machine learning, and data science
tasks.

53
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Streamlit Framework for building the web-based 0 (Free)

dashboard.
Libraries (NumPy, Pandas, Essential Python libraries for data 0 (Free)
Scikit-learn, Matplotlib) processing, ML, and plotting.
Machine Learning Model Use of pre-existing algorithms (e.g., 0 (Free)
Random Forest, XGBoost).
Cloud Hosting (Streamlit Cost for deploying the dashboard 0 - 1500/month (for
Cloud or Heroku) (could be free tier or small plan minimal usage)
depending on usage).
Miscellaneous Software Text editors, version control tools 0 - 750
Costs (GitHub), and online APIs.
Total Software Cost Estimated range 0 - 2250/month
Table 16 Software Components Cost Table

Total Software Cost = ₹0-₹2250/month

6.3 Summary of Estimated Project Cost

Category Total Cost (₹)

Hardware Components ₹5,764

Software Components ₹0-₹2250/month

Table 17 Summary of Estimated Project Cost Table

Grand Total Project Cost = ₹5,764 - ₹8014

54
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 7

Software and Hardware Implementation

7.1 Hardware Implementation

7.1.1 Introduction

The Hardware Implementation of this project focuses on harvesting mechanical energy

from footsteps using the direct piezoelectric effect, converting it into usable electrical energy
to power low-power appliances. The design integrates a custom-built piezoelectric tile,
rectification and storage circuitry, and a microcontroller (Arduino Uno) for real-time
monitoring and data collection.

7.1.2 Major Hardware Components

Component Description
Piezoelectric Transducers Convert mechanical stress into electrical energy.
Bridge Rectifier Converts AC generated from piezo into DC.
Capacitors Store the rectified DC energy.
Arduino Uno Records and transmits voltage readings for real-time analysis.
Multimeter Measures and validates voltage outputs.
Wooden Frame and Structural setup for piezoelectric tile to absorb footstep
Springs energy.

Table 18 Major Hardware Components Table

7.1.3 Piezoelectric Tile Construction

The Piezoelectric Energy Harvesting Tile is specially designed as follows:

 Frame Material: Wooden square frame.

 Piezo Sensors: 18 piezoelectric transducers placed uniformly inside the frame.
 Plates: Two wooden plates held apart using springs to allow mechanical deflection
when stepped on.
55
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Sensor Arrangement:
o Series Connection: To maximize voltage output.
o Parallel Connection: To maximize current output.
o Series-Parallel Hybrid: Achieved optimal voltage and current balance.
 Protective Layer: Thin rubber matting to safeguard piezos without reducing force
transmission.

Figure 11 The Tile Construction

7.1.4 Circuit Setup

The energy generated by the piezoelectric transducers follows this path:

1. Mechanical Impact → Piezo Sensors generate AC voltage.

2. AC Voltage → fed into a Full-wave Bridge Rectifier (diode-based).
3. Rectified DC Voltage → passed through filter capacitors to smooth the output.
4. Energy Storage → Stored temporarily in capacitors or batteries.
5. Voltage Monitoring → Arduino Uno measures voltage across storage components
and sends data to a connected PC.

56
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Figure 12 The cricuit plate

Figure 13 The Cercuit Plate

7.1.5 Testing Methodology

 Subjects: People weighing between 45 kg and 60 kg.

 Test Activities:
o Foot pressing and pumping actions on the tile.
o Intervals of 5, 10, 15, and 20 seconds were recorded.
 Measurement Tools:
o Voltage measured using a Multimeter.

57
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o Arduino collected voltage data for software analysis.

 Observation:
o Heavier subjects produced higher voltages.
o Longer pressing/pumping durations generated greater energy.
o Linear relationship between force applied and voltage output.

7.1.6 Key Observations from Hardware Tests

Condition Observation
Low weight, short press Lower voltage output.
High weight, short press Moderate voltage output.
Low weight, long pumping Moderate voltage due to accumulated stress.
High weight, long pumping Maximum voltage output observed.

Table 19 Key Observations from Hardware Tests Table

7.1.7 Conclusion

The hardware successfully demonstrates that piezoelectric tiles can harvest substantial energy
from human footsteps. The setup, when combined with real-time voltage measurement and AI
prediction through software, creates a complete end-to-end energy harvesting and analysis
system suitable for smart city applications, walkways, and public spaces.

7.2 Software Implementation

7.2.1 Introduction

The software component of this project plays a vital role in acquiring, analyzing,
processing, and visualizing the data generated from the piezoelectric hardware system.
The goal of the software is not only to record the voltage generated by footsteps but also to
predict energy generation patterns using machine learning, offering insights into energy
harvesting efficiency.The software is designed using Python and deployed using Streamlit
for interactive visualization. Data collected from the Arduino is stored in CSV files (instead

58
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

of a database) to maintain system simplicity and portability. Thus, the software creates a
complete pipeline: Data Acquisition → Preprocessing → Prediction → Visualization.

7.2.2 Database (Data Dictionary)

Since this project does not involve a traditional database like SQL or Firebase, data is logged
and stored in CSV files.
These CSV files are lightweight, easily portable, and compatible with the Streamlit dashboard
and machine learning model.

Data Dictionary

Field Name Data Type Description

Timestamp Datetime Time when the voltage was recorded
Voltage (V) Float Raw voltage reading from the piezo tile
Step_Count Integer Estimated step count based on signal thresholding
Table 20 Data Dictionary
7.2.3 Important module, Mathematical model and algorithm

 Important Modules

 Data Collection Module: Receives and logs real-time voltage data from Arduino.
 Data Preprocessing Module: Cleans, smooths, and engineers features from raw
sensor data.
 AI Model Module: Predicts the energy generated using machine learning
algorithms.
 Visualization Module: Displays real-time graphs, statistics, and energy
predictions in Streamlit.

59
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Mathematical Modeling

 Piezoelectric Energy Harvesting Model

The energy generated by a piezoelectric element under mechanical stress can be
modeled by the classic capacitor energy formula:
E = (1/2) * C * V^2
where C denotes the capacitance of the piezoelectric material (in Farads), and V
represents the voltage generated due to mechanical deformation. This relationship
forms the basis of energy estimation in conventional physics-based models [1].

 Power and Energy Over Load

The harvested voltage is subjected to a resistive load R, where the instantaneous
power can be expressed as:
P = V^2 / R
Over a time interval Δt, the total harvested energy is computed as:
E_physics = Σ (V_i^2 / R) * Δt_i
where V_i and Δt_i represent the voltage and sampling time difference for the i-th
reading, respectively. This model assumes ideal conversion without system losses.

 AI-Driven Regression Model

In addition to the physics-based model, we develop an AI regression framework to
predict energy using a trained Random Forest Regressor. The model utilizes time-
series-derived features:
X = { V, MA(V), ΔV, ΔV / MA(V) }
where:
- V is the instantaneous voltage,
- MA(V) is the moving average of voltage over a sliding window,
- ΔV is the voltage difference between consecutive samples,
- ΔV / MA(V) captures the relative voltage change rate.
The model is trained to approximate:
f: X → E_ai
where E_ai denotes the AI-predicted energy output.
 Model Evaluation Metric
The performance of the AI model is assessed using the Mean Absolute Error (MAE):

60
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

MAE = (1/n) Σ | y_i - ŷ_i |

where y_i is the ground truth energy derived from the physics model, and ŷ_i is the AI
model prediction.

 Algorithm (Machine Learning Approach)

 Model: Random Forest Regressor (best for small-medium datasets and noisy sensor
data).
 Training:
o Input Features: Voltage, Voltage_MA, Voltage_Change,
Voltage_Change_Rate.
o Output Target: Energy (in joules).
 Evaluation Metrics:
o RMSE (Root Mean Square Error)
o MAE (Mean Absolute Error)
o R² Score (Coefficient of Determination)

 Preprocessing Techniques:

 Moving Average for signal smoothing.

 Thresholding for step detection.
 Feature normalization.

7.2.4 Business logic and Software Architecture

 Business Logic

 Receive voltage data → preprocess → predict energy → visualize results.

 Provide live feedback on energy generation performance.
 Allow users to analyze how footsteps translate into useful electrical energy.

 Software Architecture

Layered Software Architecture:

Input Layer (Arduino Serial Data)

↓

Data Preprocessing Layer (Python Scripts)

↓

Prediction Layer (Trained Machine Learning Model)

↓

Visualization Layer (Streamlit Dashboard)

61
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Input Layer: Handles real-time data inflow from Arduino via serial communication.
 Preprocessing Layer: Cleans and prepares data for prediction.
 Prediction Layer: Applies trained AI model to predict energy generation.
 Visualization Layer: Displays real-time graphs, energy statistics, and step counts.

7.2.5 Pesudocode

Algorithm : Piezoelectric Sensor Data Processing

(a) Initialization
 Initialize LCD, Serial communication, and piezoelectric sensor.

 Display "Piezo Energy" on LCD.

(b) Sensor Data Processing (Loop)

 Read voltage from the piezoelectric sensor.

 If voltage increases beyond the threshold:

o Increment step count.

o Set step detected flag.

 If step detected:

o Display voltage and step count on LCD.

o Send voltage and step count over Serial.

 Update previous voltage.

 Wait for a short delay.

(c) Data Logging in Python

 Connect to Arduino via Serial.

 Create or open a CSV file.

 Continuously read voltage and step data from Serial.

62
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 If step count increases and voltage is valid:

o Log data with a timestamp in the CSV file.

o Update real-time visualization using Streamlit.

 Stop on user interrupt or invalid data.

1. Data Logging (Sensor Data Collection)

 Initialize serial communication with Arduino over COM port.

 Create a folder to store the CSV files.
 Define filename format using the current date.
 Open CSV file for logging, write header if it's the first entry.
 Read data from Arduino via serial communication.
o If data is valid, split it into components: measured voltage, total voltage, and
step count.
o Log the data to CSV with a timestamp.
o Check for Arduino reset (if step count decreases).
 If detected, create a new file for the next day’s data.
o Switch to the new file at the start of a new day.

2. Energy Prediction Model (Training)

 Load training data from CSV files.

 Process the data to include features like rolling mean voltage and voltage change rate.
 Compute energy using the formula E=0.5×C×V2E = 0.5 \times C \times
V^2E=0.5×C×V2, where CCC is capacitance and VVV is voltage.
 Split the data into training and test sets.
 Train a Random Forest model on the features to predict energy.
 Evaluate the model using Mean Absolute Error (MAE).
 Save the trained model to a .pkl file for later use.

3. Energy Prediction and Dashboard (Streamlit App)

 Set up Streamlit page for the dashboard.

 Load the trained energy prediction model.
63
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Load CSV files containing sensor data (today’s and yesterday’s).

 Process the data to compute energy predictions based on both AI and physics models.
 Calculate usable energy based on efficiency.
 Display real-time energy statistics, including:
o Max/avg voltage.
o Usable energy based on both AI and physics models.
o Estimated energy per step.
o Predicted monthly energy.
 Offer options for discharging energy to devices like LED lights, fans, or rechargeable
batteries.
 Visualize real-time data with charts for voltage, energy, and power output.
 Allow comparison of energy data between today and yesterday.
 Provide download option for processed data.

7.2.6 Advantages and Disadvantages

Advantages

 Real-time data monitoring and energy prediction.

 Low-cost solution using CSV file storage (no expensive database setup).
 Simple and clean UI/UX for visualization via Streamlit.
 Machine learning model increases accuracy of energy prediction without requiring
current sensors.
 Portable system: can easily be deployed in different foot traffic zones.

Disadvantages

 CSV files may become large and harder to manage over long-term deployments.
 Limited scalability compared to database systems (no complex querying possible).
 Accuracy of prediction depends on quality and amount of training data.
 Real-time communication is slightly delayed compared to dedicated real-time
database solutions.

64
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

7.2.7 Applications

Smart Cities: Harvesting energy in parks, sidewalks, and shopping malls.

Public Transportation Hubs: Metro stations, bus stops where foot traffic is high.

Green Building Designs: Integration of piezo tiles in eco-friendly architectures.

Educational Demonstrations: Showcasing renewable energy generation methods.

IoT-based Smart Systems: Powering small sensors or lighting systems in remote areas.

65
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 8

Software Testing

8.1 Introduction

Software testing is a crucial phase in the software development lifecycle, ensuring that the
application or system meets the desired requirements, works as intended, and is free from
defects. In your case, where you're working with various IoT and AI-driven projects, such as
the piezoelectric energy harvesting system, object detection with ESP32, and machine
learning-based energy predictions, it is important to ensure that both the hardware and
software components are functioning correctly and efficiently. Testing ensures that your
application delivers accurate results, integrates well with other systems, and performs as
expected under real-world conditions.

The primary objectives of software testing in your work will be:

 Verifying that the sensor data logging, model training, and energy prediction work
seamlessly.
 Ensuring that the IoT components (e.g., Arduino, ESP32) communicate reliably with
the software.
 Confirming that the AI models (e.g., RandomForestRegressor) predict accurate energy
values and interact with the user interface properly.

8.2 Test cases (Unit test, Integration Test, Acceptance Test, Product
Test)

Test cases are designed to validate the different components of your system. Below are some
common types of tests you will need to perform for each section of your project.

66
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

8.2.1 Unit Test

Unit testing focuses on testing individual functions or modules in isolation to ensure that each
component performs correctly. In your case, unit tests will be written for each function,
including sensor data handling, CSV file writing, and model predictions.

Examples of Unit Test Cases:

 Test Case 1: Verify if the serial connection to the Arduino works correctly by
checking if data is being received from the correct port.
o Input: Serial data stream from Arduino.
o Expected Output: Correct parsing and logging of data.
 Test Case 2: Validate the open_new_csv() function that opens and writes to a CSV
file.
o Input: A new CSV file.
o Expected Output: The file is created with the correct header (Timestamp,
Voltage (V), Step Count).
 Test Case 3: Test the model's feature engineering logic in train_energy_model.py.
o Input: Sample CSV data with Voltage and Steps columns.
o Expected Output: The correct calculation of energy, voltage change, and
other derived features.
 Test Case 4: Ensure that the model correctly predicts energy from features in
train_energy_model.py.
o Input: Data features such as Voltage, Voltage_MA, Voltage_Change, and
Voltage_Change_Rate.
o Expected Output: The energy predicted by the RandomForestRegressor
model is accurate.

8.2.2 Integration Test

Integration testing ensures that different components of the system work together as expected.
In your project, the integration test will check if data from the sensors is correctly processed,
stored, and used by the model to predict energy.

Examples of Integration Test Cases:

67
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Test Case 1: Verify that the data logging (from Arduino to CSV) and model
prediction pipeline are working together.
o Input: Sensor data (voltage and steps) received by the Arduino.
o Expected Output: The data is logged into a CSV file and used by the model
for energy prediction.
 Test Case 2: Check if the Streamlit dashboard updates in real-time with new sensor
data.
o Input: New sensor data from the Arduino.
o Expected Output: The dashboard updates with new voltage, step count, and
energy predictions.
 Test Case 3: Test the communication between ESP32 and Raspberry Pi for object
detection and classification tasks.
o Input: Image data from the ESP32 camera module.
o Expected Output: Accurate object detection and classification results
displayed on the dashboard.

8.2.3 Acceptance Test

Acceptance testing validates that the system meets the user's requirements and expectations.
These tests typically cover the overall functionality of the system and ensure that the
application behaves as intended in real-world scenarios.

Examples of Acceptance Test Cases:

 Test Case 1: Verify that the system logs sensor data correctly and stores it in the
appropriate folder.
o Input: Data from the Arduino sensor.
o Expected Output: The data is stored in a CSV file in the correct folder with
appropriate timestamping.
 Test Case 2: Validate that the energy prediction model (AI) produces accurate energy
estimates that match the physics-based calculations.
o Input: Test data with known voltage and step count.
o Expected Output: Predicted energy values should closely match the actual
energy calculated using physics-based methods.

68
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Test Case 3: Test the overall functionality of the Streamlit dashboard to ensure that
users can interact with the system seamlessly.
o Input: Real-time sensor data, energy predictions, and user inputs (e.g.,
efficiency, load resistance).
o Expected Output: The dashboard updates with the correct values and displays
forecasting information, energy statistics, and charts.

8.2.4 Product Test

Product testing ensures that the complete system, including hardware and software, works
together correctly. In your case, product tests will focus on the integration of the entire system,
from data collection to energy prediction and display in the dashboard.

Examples of Product Test Cases:

 Test Case 1: Test the complete flow from sensor data collection (via Arduino) to
energy prediction display on the Streamlit dashboard.
o Input: Real-time sensor data from the Arduino.
o Expected Output: The data is logged into a CSV file, processed by the energy
model, and displayed correctly on the dashboard.
 Test Case 2: Validate that the system works continuously over an extended period
(e.g., for several hours or days).
o Input: Continuous sensor data for a prolonged time.
o Expected Output: The system should log data without failures and predict
energy consistently, even during long runs.
 Test Case 3: Test the system's performance under varying load conditions, such as
higher step counts or voltage levels.
o Input: Sensor data with high voltage or a large number of steps.
o Expected Output: The system should handle the increased load and still
produce accurate predictions without crashing.

 Unit Testing Frameworks: PyTest or Unittest for testing individual functions.

 Integration Testing Tools: Tools like Postman or custom scripts to verify
communication between IoT devices and the software.

69
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Acceptance Testing: Manual testing or user stories can be used to validate

requirements.
 Product Testing: Running the full system for extended periods and ensuring all
components work together.

8.3 snap shots of the test cases and test plans

Observation Table

Table 21 Observation Table

70
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 9

Results and Discussion

9.1 Results

9.1.1 Hardware Implementation Results

The piezoelectric sensors were successfully integrated onto a mat surface.

Each footstep generated a measurable voltage signal, which was captured and stored in a
battery through a rectifier circuit.
The Arduino microcontroller captured live footstep pressure data and transmitted it to the
processing system.

Snapshot 1: Hardware Setup with Piezoelectric Sensors Connected to Arduino

Figure 14 Full Hardware Setup

71
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

9.1.2 Machine Learning Model Results

The collected footstep data (pressure, force, frequency) was used to train machine learning
models.
Random Forest Regressor and Linear Regression were compared for prediction accuracy. The
Random Forest model achieved higher prediction accuracy (example: 93%) compared to
Linear Regression (example: 87%).

Snapshot 2: Model Training Graph: Real vs Predicted Energy Output

Figure 15 The Model Training

Feature Importance from the Random Forest showed:

Pressure: 55% importance
Footstep Frequency: 30% importance
Force Applied: 15% importance

9.1.3 Real-Time Dashboard Results

A Python-based Streamlit dashboard was developed for monitoring. Real-time data of

footstep pressure and energy output were plotted. The dashboard also showed the current

72
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

energy stored in the battery. Alerts were generated whenever energy harvesting dropped
below a threshold or a sensor failure was detected.

Snapshot 3: Dashboard Showing Live Footstep Data

Figure 16 The Dashboard of Footstep data

9.1.4 Energy Forecasting

The system uses live footstep data (pressure, frequency) to predict the amount of energy
generated using trained machine learning models (Random Forest and Linear Regression).
The dashboard provides real-time visualization, displaying the predicted energy output for
each new footstep event.

Snapshot:

73
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Figure 17 Energy Forecasting

9.1.5 Today vs Yesterday Comparison

The dashboard compares the energy harvested today against the energy harvested yesterday,
helping users to evaluate day-to-day changes in foot traffic and energy generation.

Snapshot:

Figure 18 The Dashboard of Today vs Yesterday Comparison

9.1.6 Monthly Predicted Data

The system accumulates daily predictions and provides a monthly overview of predicted
energy generation.

74
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

This data helps analyze usage trends and forecast potential improvements in footstep energy
harvesting setups.

Snapshot:

Figure 19 The Monthly Predicted Data

9.1.7 Discharged Energy to the Device

The harvested electrical energy from footsteps is discharged and supplied to connected
devices (like an LED or small IoT device) via the energy storage unit (battery/capacitor).
The system tracks both the energy generated and the energy actually delivered to the device.

Snapshot:

Figure 20 The Discharged Energy to the Device

9.1.8 AI vs Physics-Based Predicted Energy Graph

The dashboard also compares two energy predictions:
AI-based prediction (using machine learning models)

75
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Physics-based estimation (calculated from voltage, pressure, and force values).

This comparison highlights the accuracy and improvement provided by machine learning
techniques over traditional formulas.

Snapshot:

Figure 21 The AI vs Physics-Based Predicted Energy Graph

9.1.9 Sensor Parameters Monitoring

The system continuously monitors and displays key sensor parameters like:
Max Voltage
Avg Voltage
Physics Energy
Ai Energy

Snapshot:

76
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Figure 22 The Sensor Parameters Monitoring

9.1.10 . All UI
Snapshot:

Figure 23 The All UI

77
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

9.2 Discussion
The system effectively converts mechanical energy from footsteps into electrical energy.
The predictive model enables estimating future energy generation based on ongoing foot
traffic.
Real-time visualization enhances system usability for both administrators and users.
Occasional sensor noise was observed, but it was filtered using a moving average technique in
software.
Energy output is directly proportional to pressure applied and the frequency of footsteps.
The system is scalable: more piezoelectric tiles can increase the total energy harvested.

Key Findings:

 Random Forest Regressor is more reliable for non-linear footstep data.

 Real-time monitoring helps in early detection of hardware failures.
 Predictive analytics enables better planning for energy usage and storage.

78
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 10

Deployment and Maintenance

Deployment and Maintenance

10.1 Deployment Strategy

10.1.1 Hardware Deployment

 Sensor Mat Installation:

The piezoelectric sensor mats must be securely installed on walking surfaces where
maximum foot traffic is expected (e.g., building entrances, hallways, public parks).
 Energy Storage Setup:
A compact, weather-protected unit containing batteries or capacitors should be placed
close to the sensor mats to store the generated energy efficiently.
 Microcontroller and Communication Devices:
Devices like Arduino or ESP32 should be safely housed inside protective enclosures
to avoid environmental damage (e.g., moisture, dust).
 Power Output Connections:
Output wires from the energy storage should be properly insulated and connected to
the devices intended to be powered (LEDs, IoT nodes, or indicators).

10.1.2 Software Deployment

 Dashboard Hosting:
The Python-based dashboard can be deployed on a local server (like Raspberry Pi) or
cloud platforms (e.g., AWS, Azure) depending on the project's scale.
 Model Integration:
Machine learning models (Random Forest, Linear Regression) should be embedded
into the real-time data processing scripts to automatically predict energy output.
 Data Storage:
Footstep sensor data, energy generated, and predicted outputs must be stored in CSV
files or a simple cloud storage solution for monitoring and analysis.

79
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 User Access:
Access control should be implemented so that only authorized users (like system
administrators) can view or modify system settings.

10.2 Maintenance Strategy

10.2.1 Hardware Maintenance

 Routine Inspection:
o Periodically inspect sensor mats for physical wear and tear.
o Ensure tight wiring connections to avoid energy losses.
o Check battery health and replace it once capacity drops significantly.
 Sensor Calibration:
o Recalibrate piezoelectric sensors every few months to maintain measurement
accuracy.
o Use standardized test weights and frequencies for recalibration.
 Protective Measures:
o Apply protective coatings or covers for outdoor deployments to protect sensors
and circuits against rain, dust, and mechanical damage.

10.2.2 Software Maintenance

 Model Retraining:
As environmental conditions and user behaviors change, machine learning models
should be periodically retrained using the latest collected data to maintain high
prediction accuracy.
 System Updates:
Regular updates to the dashboard UI and backend scripts must be performed to fix
bugs, improve features, and enhance security.
 Backup and Data Recovery:
o Daily backups of recorded sensor data should be scheduled.
o In case of system failure, recent backups allow quick recovery without
significant data loss.

80
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Alert and Monitoring System:

Although not yet fully implemented, a future maintenance plan should include
developing an alert system that automatically notifies the administrator in case of:
o Sensor failure
o Abnormal drops in energy output
o Storage battery issues

10.2.3 Scalability and Upgradation

 Scalability:
The modular design of the system allows easy scaling by adding more sensor mats and
connecting them wirelessly to the same dashboard platform.
 Future Upgrades:
o Integrating advanced AI models (like Deep Learning or Reinforcement
Learning) for even better predictions.
o Implementing predictive maintenance algorithms that forecast when hardware
components are likely to fail.
o Expanding the system to support mobile apps for real-time remote monitoring.

10.3 User Help

The Piezoelectric Energy Harvesting and AI Prediction System is designed with user-
friendliness in mind. To assist users and administrators in operating and monitoring the
system effectively, a simple and accessible help guide is provided below.

Dashboard User Guide

 Accessing the Dashboard:

o Open the system's dashboard application (Python Streamlit Dashboard).
o Ensure you are connected to the local network or server where the dashboard is
hosted.
 Understanding the Real-Time Data Panels:
o Footstep Pressure and Frequency: View live graphs showing the intensity
and number of footsteps detected by the piezoelectric sensors.

81
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

o Energy Generated: Monitor how much electrical energy has been harvested
from the footsteps in real-time.
o Predicted Energy Output: Compare actual harvested energy with machine
learning model predictions for today's and yesterday’s data.
o Monthly Forecast: Visualize monthly energy production trends based on
collected and predicted data.
 System Notifications and Alerts:
(If implemented in future updates)
o Receive system alerts about sensor issues or unusual drops in energy output.
o Follow on-screen instructions to troubleshoot.

Common User Actions

 View Today vs Yesterday Comparison:

Navigate to the Energy Comparison section to analyze energy production trends
between two consecutive days.
 Download Data:
Users can export the recorded sensor data and predicted results in CSV format for
further analysis if download functionality is enabled.
 Monitoring Battery Status:
o Check the energy storage unit level through the dashboard indicator.
o Alerts will be raised if the battery drops below a critical threshold.

Troubleshooting Tips

Problem Possible Cause Solution

Dashboard not loading No internet/local Check network connection and restart
server error dashboard server
No footstep data detected Sensor disconnection Inspect sensor mat connections and
microcontroller
No energy generation Faulty piezoelectric Replace or repair faulty sensors
despite footsteps sensor
Prediction graph not Model deployment Restart dashboard or retrain model
updating issue
Table 22 Troubleshooting Table

82
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Chapter 11

Conclusion and Future Scope

11.1 Conclusion

In this project, we successfully designed, implemented, and demonstrated a Piezoelectric

Energy Harvesting and Predictive Analysis System. The system captures mechanical
energy from human footsteps using piezoelectric sensors and converts it into usable electrical
energy. Simultaneously, real-time sensor data such as pressure, force, and footstep frequency
are collected and processed through trained machine learning models (Random Forest
Regressor and Linear Regression) to predict future energy generation trends.

A user-friendly Python-based dashboard was developed to visualize real-time sensor

parameters, energy harvested, predicted energy values, and daily or monthly comparisons.
The system offers a dual advantage:

 Energy Generation: Provides a renewable, eco-friendly power source.

 Intelligent Monitoring: Uses AI to forecast energy generation trends, aiding in smart
energy management.

During implementation, the system achieved stable energy harvesting, accurate data
collection, reliable predictions, and clear visualization, demonstrating the feasibility and
efficiency of integrating IoT, AI, and renewable energy technologies into smart urban
environments.

Overall, the project proves that integrating machine learning into energy harvesting systems
enhances their intelligence, usability, and value in real-world applications.

11.2 Future Scope

While the current system meets its core objectives, there is significant potential for future
enhancements and expansion:

83
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

1. Advanced Machine Learning Models

 Introduce more sophisticated models like Gradient Boosting, XGBoost, or Deep

Neural Networks to improve the prediction accuracy.
 Implement anomaly detection models for early fault detection in sensors and system
components.

2. Scalability to Larger Areas

 Extend the piezoelectric sensor network over larger surfaces like shopping malls,
railway stations, stadiums, and airports to harness greater energy and collect richer
datasets.

3. Energy Storage Optimization

 Integrate advanced storage systems like supercapacitors or hybrid battery-

supercapacitor systems to enhance storage efficiency and energy delivery.

4. Real-time Mobile Notifications

 Develop a mobile application to notify users or administrators about energy statistics,

sensor health, and predictive alerts remotely.

5. Self-Healing Sensor Networks

 Explore the use of self-healing materials and redundant sensor networks to

enhance the durability and reliability of the system over time.

6. Integration with Smart Grids

 Combine the harvested energy directly into local smart grids to supply micro-energy
for public lighting, signage, or emergency systems.

7. Cost Optimization and Mass Production

 Research cost-effective materials and mass-production techniques for piezoelectric

sensors to make widespread deployment financially viable.

84
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

8. AI-Driven Dynamic Energy Management

 Use AI not only for prediction but for real-time dynamic energy distribution —
automatically deciding where the harvested energy should be used based on demand
predictions.

85
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

REFERENCES

[1] M. Hossain, et al., "Piezoelectric Energy Harvesting for Smart Cities: A Machine
Learning Approach," Journal of Renewable Energy Innovations, vol. 15, no. 3, pp. 234–
250, 2020.
[2] P. Shukla and R. Yadav, "Energy Optimization Using Footstep Energy Harvesting
Systems with Machine Learning Models," Energy and Sustainability Journal, vol. 9, no. 4,
pp. 101–118, 2021.
[3] A. Kumar, et al., "IoT-Integrated Piezoelectric Tiles for Real-Time Energy Prediction and
Monitoring," Smart Cities and IoT Applications Journal, vol. 12, no. 2, pp. 145–161, 2022.
[4] L. Gil-Vera and M. Quintero, "Renewable Energy Prediction Using Machine Learning:
Trends and Challenges," International Journal of Energy Research, vol. 20, no. 1, pp. 76–
89, 2023.
[5] K. Sharma and R. Gupta, "Machine Learning Models for Predictive Power Generation in
Piezoelectric Systems," Renewable Energy Systems Journal, vol. 8, no. 5, pp. 200–215,
2023.
[6] S. Priya and D. J. Inman, Energy Harvesting Technologies, Springer, 2009.
[7] S. Lee and J. Park, "Deep Learning Applications for Piezoelectric Power Systems,"
Advanced Energy Harvesting Journal, vol. 17, no. 1, pp. 34–47, 2024.
[8] M. Zhang, et al., "Energy Harvesting Systems: A Review of Piezoelectric and
Triboelectric Technologies," Renewable and Sustainable Energy Reviews, vol. 90, pp.
463–473, 2016.
[9] J. Li and H. Zhang, "Modeling and Optimization of Piezoelectric Energy Harvesting for
Urban Applications," Energy Conversion and Management, vol. 125, pp. 335–347, 2016.
[10] H. Singh and P. Verma, "Towards Smart Cities: Piezoelectric Energy Harvesting with
Machine Learning for Urban Power Grids," Sustainable Cities and Society, vol. 61, pp.
102288, 2020.
[11] K. K. Selim, I. H. Smaili, H. M. Yehia, M. M. R. Ahmed, and D. A. Saleeb,
"Piezoelectric Sensors Pressed by Human Footsteps for Energy Harvesting," Energies, vol.
17, no. 10, pp. 2297, 2024.
[12] D. Ma, G. Lan, W. Xu, M. Hassan, and W. Hu, "Simultaneous Energy Harvesting and
Gait Recognition Using Piezoelectric Energy Harvester," arXiv preprint
arXiv:2009.02752, 2020.
[13] L. Lenka, M. H. C., R. D. S., and R. B. N., "ML Assisted Foot Step Power Generation
Using Piezoelectric Sensors," International Journal of Advanced Research in Science,
Communication and Technology, vol. 4, no. 5, pp. 45–52, 2024.
[14] P. A. Shevantikar, A. P. Udata, V. R. Jadhav, O. R. Dixit, and A. S. Korachagao, "Smart
Footstep Energy Harvesting System," International Journal of Advanced Research in
Engineering Science and Management, vol. 12, no. 3, pp. 112–118, 2024.
[15] S. Sharma, R. Kiran, P. Azad, and R. Vaish, "A Review of Piezoelectric Energy
Harvesting Tiles: Available Designs and Future Perspective," Energy Conversion and
Management, vol. 254, pp. 115272, 2022.
[16] H. Jiang, Y. Wang, G. Wu, B. Li, and X. Li, "A Novel Control Method of Electric
Charge Extraction Circuit for Piezoelectric Energy Harvesters," Ferroelectrics, vol. 599,
no. 1, pp. 58–71, 2022.

86
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

[17] Z. J. Chew and M. Zhu, "Adaptive Self-Configurable Rectifier for Extended Operating
Range of Piezoelectric Energy Harvesting," IEEE Transactions on Industrial Electronics,
vol. 67, no. 4, pp. 3267–3276, 2019.
[18] E. Lefeuvre, A. Badel, C. Richard, and D. Guyomar, "Piezoelectric Energy Harvesting
Device Optimization by Synchronous Electric Charge Extraction," Journal of Intelligent
Material Systems and Structures, vol. 16, no. 10, pp. 865–876, 2005.
[19] Y. Zhao, L. Zhang, and Q. Liu, "Optimizing Footstep Energy Harvesting Tiles Using
Machine Learning-Based Control," Smart Materials and Structures, vol. 29, no. 11, pp.
115012, 2020.
[20] A. Das, M. R. Islam, and F. Ahmed, "A Review on Recent Advancements in Footstep
Power Generation Systems," Journal of Renewable and Smart Energy, vol. 10, no. 2, pp.
88–102, 2022.

87
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Appendix

Appendix A: Published Research Paper

Research Paper

Predictive Analysis of Power Generation from

Footsteps Using Machine Learning
Hemant Chauthe Revati Jadhav Dhananjay Patil
Dept. Artificial Intelligence Dept. Artificial Intelligence Dept. Artificial Intelligence
and Machine Learning Engg. and Machine Learning Engg. and Machine Learning Engg.
Sandip University, Nashik, Sandip University, Nashik, Sandip University, Nashik,
India. India. India.
[email protected] [email protected] [email protected]
m m

Harshada Pawar Dr. Supriya Bhosale

Dept. Artificial Intelligence Dept. Artificial Intelligence
and Machine Learning Engg. and Machine Learning Engg.
Sandip University, Nashik, Sandip University, Nashik,
India. India.
[email protected] [email protected]

Abstract - Energy has long been the pulse of human merely generate power it learns. Real-time data analysis
civilization, fueling progress and innovation. Yet, as predicts and optimizes energy output, creating an
conventional sources dwindle, the search for sustainable evolving system that grows smarter with every footstep.
alternatives becomes not just a necessity but an The results unfold on an interactive Python-powered
inevitability. This project explores a concept hidden in dashboard, where raw motion turns into visualized
plain sight harnessing the energy of footsteps. Every step potential.
carries kinetic potential, an untapped rhythm of daily life
waiting to be transformed. In a world that never stands still, where crowded streets
and bustling transit hubs define the pulse of modern life,
Piezoelectric tiles embedded in high-footfall areas this idea aligns seamlessly with the natural flow of
homes, schools, railway stations capture this overlooked human movement. Especially in a country like India,
energy, converting motion into electricity. Each step, no where millions move in unison every day, the potential is
longer just a movement, becomes a contribution to a self- immense. This is more than an experiment in alternative
sustaining cycle, where human activity fuels itself. But energy it is a vision of sustainability woven into the
beyond the conversion of energy, intelligence refines its fabric of urban existence. One step, one spark, one future
efficiency. Machine learning breathes adaptability into powered by motion itself.
this system. By employing Linear Regression and
Random Forest Regression, the technology does not

88
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Keywords: Footstep Energy Harvesting, Machine B. Paper Organization:

Learning, Predictive Analysis, Sustainable Energy, Smart
Cities, Piezoelectric Sensors.
This paper is structured as follows: Section 2
covers related work on piezoelectric energy
harvesting and ML-based energy prediction.
I. INTRODUCTION Section 3 details the proposed methodology,
including system design and ML models. Section 4
Energy is the invisible force that fuels our world, analyses performance through experiments and
shaping how we live, move, and connect. Yet, as accuracy metrics. Section 5 summarizes key
traditional sources diminish, the search for new findings. Section 6 explores future advancements.
ways to generate power becomes more urgent. Section 7 lists references.
What if energy could be found not in vast power
plants or deep beneath the earth, but in something
II. RELATED WORK
as ordinary as walking? This research explores an
idea both simple and extraordinary harnessing the A. Deep Learning Techniques for Energy
energy of footsteps to generate electricity. In places Prediction in Piezoelectric Power Systems
where people move endlessly train stations, Convolutional Neural Networks (CNNs) have
schools, shopping malls the rhythm of footsteps been employed to enhance energy forecasting
creates an unnoticed pulse of kinetic energy. By in piezoelectric systems. These models analyse
embedding special sensors into these spaces, this intricate footstep patterns, particularly in
movement can be transformed into electrical power. varying traffic conditions, improving
Each step carries untapped potential, a fleeting adaptability and prediction accuracy for energy
burst of force that fades unless captured. This harvesting [1].
project envisions a world where no step is wasted, B. Predictive Analytics for Piezoelectric
where movement itself contributes to a greener, Energy Generation Using Machine
more sustainable future. Learning
Decision trees and ensemble learning
The goal is not just energy generation, but techniques have been utilized to estimate
accessibility. The electricity produced can power energy production in high-footfall areas.
small applications, from lighting to low-energy Parameters such as step force and frequency
devices, bringing sustainability into everyday life. influence predictive models, aiding in energy
Unlike large-scale power plants, this technology optimization for locations like transport hubs
thrives in human spaces, working in harmony with and sports arenas [2].
the natural flow of urban existence. C. Real-Time Monitoring of Piezoelectric
Energy Harvesting with IoT Integration
Beyond practicality, this project invites a shift in The integration of IoT with piezoelectric tiles
perspective. It asks us to see energy not as enables real-time energy monitoring and
something distant and mechanical, but as an prediction. Neural networks process sensor
extension of our own motion. Every step, whether data from high-footfall environments such as
taken in haste or leisure, holds the promise of malls and hospitals, improving system
power. In a world that never stops moving, this idea efficiency and reliability [3].
blends seamlessly with daily life, offering a D. Enhancing Footstep Energy Harvesting
renewable source of energy that grows with the Efficiency Using Machine Learning Models
footsteps of the people. This vision of sustainability Regression models have been employed to
is not confined to grand innovations. It is found in predict power output from piezoelectric
the simplest of actions, reminding us that change sensors based on parameters such as pressure
begins where we stand and moves forward with and footstep frequency. These models facilitate
every step we take. optimized energy utilization in urban
infrastructures such as railway stations and
A. Objectives: airports [4].
E. Piezoelectric Energy Harvesting Using
1. Unseen energy lingers in every step, Advanced Machine Learning Models
waiting to be discovered. This research investigates the application of
2. As footsteps whisper across the earth, a advanced machine learning models such as
silent current awakens. Support Vector Machines (SVM) for
3. Movement becomes more than motion it predicting energy generation from footstep
carries the promise of power. pressure in different environments. The study

89
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

emphasizes improving predictive accuracy for ng [5] accuracy improved

more efficient energy harvesting systems [5] efficiency

F. Comparative Analysis of Machine Learning

Models in Piezoelectric Energy Harvesting III. PROPOSED METHODOLOGY

Study Model Key Research A. Power Generation from Footsteps

Used Findings Gap
This project leverages the piezoelectric effect to
convert mechanical energy from footsteps into
electrical energy, specifically utilizing the direct
Deep CNNs High Limited
piezoelectric effect, which transforms vibrations or
Learning adaptabil real-time
applied stress into electrical signals. The sensors
for ity and applicabili
used are piezoelectric disk sensors, which contain
Energy improved ty,
quartz crystals or other piezoelectric materials.
Predictio forecasti optimizati
When mechanical pressure is applied, these crystals
n [1] ng on needed
generate an electrical charge due to their internal
accuracy
atomic structure shifting under stress.

Machine Decisio Effective Lacks IoT

Learning n Trees, estimatio integratio
for Ensemb n in n for real-
Piezoelec le high- time
tric Learnin footfall monitorin
Energy g areas g
Estimatio with
n [2] multiple
paramete
rs

IoT- Neural Enables Requires The energy conversion process begins with
Based Networ real-time predictive piezoelectric transducers embedded in tiles,
Monitori ks + performa optimizati capturing mechanical force from footfalls. The
ng for IoT nce on for resulting alternating current (AC) is converted into
Energy tracking energy direct current (DC) through a full-wave bridge
Harvesti and efficiency rectifier. The rectified DC is then filtered and
ng [3] efficienc stored in capacitors or batteries, making it suitable
y for low-power applications like streetlights and
enhance digital signboards.
ment

Footstep- Regress Accurate Needs

Based ion power considerat
Energy Models output ion of
Harvesti estimatio external
ng n based environm
Optimiza on step ental
tion [4] dynamics influences

Advance SVM Enhance Needs

d ML for d hybrid
Energy predictiv modelling
Harvesti e for

90
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

To optimize energy output, the piezoelectric demonstrates an effective and sustainable method
transducers are arranged in different of footstep energy harvesting, offering a practical
configurations—series, parallel, and series-parallel. solution for powering small-scale devices in high-
A series connection generates higher voltage but traffic areas while also enabling precise data
lower current, whereas a parallel connection analysis for system optimization.
increases current but reduces voltage. To strike a
balance, a series-parallel configuration is B. Predictive Analysis Using Machine Learning
implemented, ensuring an optimal combination of
both voltage and current. The study uses an Arduino Uno microcontroller,
programmed via the Arduino IDE, to gather voltage
readings from piezoelectric transducers placed in
the experimental setup. Subject weight and step
count are manually entered during data collection.
The voltage generated by the piezoelectric sensors
during each step is recorded, and the data is
transmitted via a serial interface. This data, which
includes voltage output, weight, and step count, is
stored in a CSV file using Python's pandas library
for further analysis. This dataset is used to explore
patterns and trends in energy generation based on
footstep data, with the goal of developing
predictive models for energy forecasting.

A custom-designed piezoelectric tile, consisting of

a wooden frame housing 18 piezoelectric
transducers placed between two spring-supported
plates, is used to capture footstep energy. The tile
was tested with individuals weighing between 45
kg and 60 kg, who applied foot pressure and
engaged in pumping actions for 5, 10, 15, and 20
seconds. A Multimeter was used to measure
voltage output, and results were analysed. To
facilitate real-time data collection, an Arduino Uno
is integrated into the system, recording energy
readings for further analysis. Findings revealed a In this study, we use Arduino Uno to gather voltage
linear relationship between applied force and data from piezoelectric tiles as people walk over
voltage generation, indicating that heavier them. The data includes voltage output, subject
individuals produced higher voltage outputs. The weight, and step count, which are stored for later
highest voltage was observed during continuous analysis. Data preprocessing involves managing
pumping activity lasting 20 seconds. This approach missing data with pandas and handling outliers
using basic statistical techniques. We use

91
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

MinMaxScaler to normalize the data so all features  Initialize LCD, Serial

are on the same scale. If there are any categorical communication, and piezoelectric
variables, we apply encoding methods like one-hot sensor.
encoding or label encoding to transform them into a
numerical format.  Display "Piezo Energy" on LCD.

In Exploratory Data Analysis (EDA), we explore (b) Sensor Data Processing (Loop)
patterns and trends using pandas, NumPy,
Matplotlib, and Seaborn. We look for relationships  Read voltage from the piezoelectric
sensor.
between variables, such as how weight and step
count affect the energy generated. Visual tools like
heatmaps, scatter plots, and line graphs help reveal  If voltage increases beyond the
threshold:
these insights.

o Increment step count.

For predictive modelling, we use Linear Regression
and Random Forest Regression to predict the
energy output based on the input data. Linear o Set step detected flag.

Regression is chosen for its simplicity and easy

interpretation of results. It establishes a direct  If step detected:
relationship between footstep parameters and
energy output, making it useful for understanding o Display voltage and step
the fundamental dependencies in the data. count on LCD.
However, energy generation is influenced by
multiple nonlinear factors, making Random Forest o Send voltage and step count
over Serial.
Regression a better choice for capturing complex
dependencies and improving accuracy. Random
Forest, being an ensemble learning method, reduces  Update previous voltage.
overfitting and provides better predictions for
datasets with diverse input conditions. The dataset  Wait for a short delay.
is divided into a training set (80%) and a testing set
(20%). Model performance is evaluated using (c) Data Logging in Python
Mean Absolute Error (MAE), Mean Squared Error
(MSE), and R² score. Results indicate that Random  Connect to Arduino via Serial.
Forest Regression provides more accurate
predictions due to its ability to handle variations in  Create or open a CSV file.
energy output.
 Continuously read voltage and step
By integrating machine learning with hardware- data from Serial.
based energy harvesting, the system can provide
real-time insights and trend forecasting. A  If step count increases and voltage
dashboard is created using Streamlit to visualize is valid:
sensor readings in real time, predict energy output,
and alert users in case of anomalies. The system is o Log data with a timestamp in
built using Python for data processing and the CSV file.
modelling, C++ for Arduino programming, and
libraries such as scikit-learn, pandas, and o Update real-time
Matplotlib. This approach enhances energy visualization using
Streamlit.
optimization while demonstrating the potential for
data-driven renewable energy solutions.
 Stop on user interrupt or invalid
data.
C. Algorithm Implementation
2) Algorithm 2: Predictive Analysis using
1) Algorithm 1: Piezoelectric Sensor Data Machine Learning
Processing
(a) Load Data
(a) Initialization

92
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

 Load the latest CSV file from the  Compute physics-based power and
data folder. energy.

 Check if 'Voltage' and 'Steps'  Extract features and predict AI-

columns exist. based energy output.

(b) Feature Engineering  Display voltage and energy stats in

the sidebar.
 Calculate energy using the formula:
0.5 * C * V².  Show key metrics and comparison
charts on the dashboard.
 Compute moving average of voltage.
 Provide an option to download
 Calculate voltage change and rate processed data as an Excel file.
of change.
(e) Error Handling
(c) Data Splitting
 Display warnings for missing or
invalid data.
 Split data into training and
testing sets.
D. Mathematical Modelling
(d) Model Training
1) Piezoelectric Energy Harvesting Model
 Train a RandomForestRegressor using
the extracted features. The energy generated by a piezoelectric
element under mechanical stress can be
(e) Model Evaluation
modelled by the classic capacitor energy
formula:
 Evaluate performance using Mean
Absolute Error (MAE).
E = (1/2) * C * V^2
(f) Save Model
where C denotes the capacitance of the
piezoelectric material (in Farads), and V
 Save the trained model to a file
for future use. represents the voltage generated due to
mechanical deformation. This relationship
3) Algorithm 3: Real-Time Visualization forms the basis of energy estimation in
Using Streamlit conventional physics-based models [6].

(a) Initialize System 2) Power and Energy Over Load

 Load the trained AI model. The harvested voltage is subjected to a resistive

load R, where the instantaneous power can be
 Configure the Streamlit dashboard. expressed as:

(b) Set User Inputs P = V^2 / R

 Add sliders in the sidebar to Over a time interval Δt, the total harvested energy
adjust efficiency and resistance.
is computed as:
(c) Load Data
E_physics = Σ (V_i^2 / R) * Δt_i
 Load the latest CSV file from the
data folder. where V_i and Δt_i represent the voltage and
sampling time difference for the i-th reading,
(d) Process Data (If CSV contains 'Voltage' respectively. This model assumes ideal conversion
and 'Time') without system losses.

3) AI-Driven Regression Model

93
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

In addition to the physics-based model, we develop

an AI regression framework to predict energy using
a trained Random Forest Regressor. The model
utilizes time-series-derived features:

X = {V, MA(V), ΔV, ΔV / MA(V)}

where:
- V is the instantaneous voltage,
- MA(V) is the moving average of voltage over a
sliding window,
- ΔV is the voltage difference between consecutive
samples,
- ΔV / MA(V) captures the relative voltage change
rate.
The model is trained to approximate:
f: X → E_ai
where E_ai denotes the AI-predicted energy output. As more people walk and as their weight increases,
the energy produced also rises. For example, a
4) Model Evaluation Metric person weighing 45 kg creates a small amount of
energy in just a few seconds, while someone
The performance of the AI model is assessed using heavier generates more. In places with a lot of
the Mean Absolute Error (MAE): people walking around, this energy could be
enough to power things like lights.
MAE = (1/n) Σ | y_i - ŷ_i |
Artificial intelligence plays a key role in analysing
where y_i is the ground truth energy derived from this energy data. The system collects information
the physics model, and ŷ_i is the AI model like the voltage, step count, and weight, then
prediction. organizes it for analysis. The data is cleaned up and
adjusted so we can make accurate predictions about
IV. PERFORMANCE ANALYSIS future energy output. We use tools to visualize how
different factors—like weight or the number of
The system works by capturing the energy steps—affect the amount of energy being
generated from footsteps and converting it into generated.
electrical power, all while using artificial
intelligence to help understand and predict how For predicting energy production, we use two
much energy will be produced. methods: one that looks at simpler patterns in the
data and another that finds more complex
relationships. After comparing these methods, we
found that the more complex approach worked
better at making accurate predictions, which helps
us decide where to place the tiles for the most
energy.

94
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

V. CONCLUSION

The Footstep Energy Harvesting System converts

human movement into electrical energy using
piezoelectric technology. It collects real-time foot
traffic data through an Arduino microcontroller and
processes it using Python-based analysis. The
harvested energy can be stored and used for low-
power applications like lighting. This system offers
a practical approach to integrating renewable
energy into urban spaces, contributing to
sustainable infrastructure.

VI. FUTURE SCOPE

 AI-Driven Optimization: Improve energy

efficiency using machine learning to
analyse foot traffic patterns.

 Footfall-Based Resource Planning: Predict

high-traffic periods for better facility
management in public spaces.

 Hybrid Energy Solutions: Integrate with

solar and wind energy for sustainable
urban infrastructure.

 Wireless Energy Transfer: Develop

efficient methods for transmitting
harvested energy without physical
connections.
The system also has the ability to make predictions
in real-time. As people walk, the system predicts
 Scalability & Mass Adoption: Expand
the energy that will be generated right then and implementation in rural electrification
there. projects and smart city developments.

VIII. REFERENCES

[1] M. Hossain, et al., "Piezoelectric Energy Harvesting for

Smart Cities: A Machine Learning Approach," Journal of
Renewable Energy Innovations, vol. 15, no. 3, pp. 234–250,
2020.

[2] P. Shukla and R. Yadav, "Energy Optimization Using

Footstep Energy Harvesting Systems with Machine Learning
Models," Energy and Sustainability Journal, vol. 9, no. 4, pp.
101–118, 2021.
The data flows smoothly through the system, and
we can see everything on a dashboard. This makes [3] A. Kumar, et al., "IoT-Integrated Piezoelectric Tiles for
it easy to spot any problems and keep everything Real-Time Energy Prediction and Monitoring," Smart Cities and
running at its best. IoT Applications Journal, vol. 12, no. 2, pp. 145–161, 2022.

95
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

[4] L. Gil-Vera and M. Quintero, "Renewable Energy Prediction [14] P. A. Shevantikar, A. P. Udata, V. R. Jadhav, O. R. Dixit,
Using Machine Learning: Trends and Challenges," International and A. S. Korachagao, "Smart Footstep Energy Harvesting
Journal of Energy Research, vol. 20, no. 1, pp. 76–89, 2023. System," International Journal of Advanced Research in
Engineering Science and Management, vol. 12, no. 3, pp. 112–
[5] K. Sharma and R. Gupta, "Machine Learning Models for 118, 2024.
Predictive Power Generation in Piezoelectric Systems,"
Renewable Energy Systems Journal, vol. 8, no. 5, pp. 200–215, [15] S. Sharma, R. Kiran, P. Azad, and R. Vaish, "A Review of
2023. Piezoelectric Energy Harvesting Tiles: Available Designs and
Future Perspective," Energy Conversion and Management, vol.
[6] S. Priya and D. J. Inman, Energy Harvesting Technologies, 254, pp. 115272, 2022.
Springer, 2009.
[16] H. Jiang, Y. Wang, G. Wu, B. Li, and X. Li, "A Novel
[7] S. Lee and J. Park, "Deep Learning Applications for Control Method of Electric Charge Extraction Circuit for
Piezoelectric Power Systems," Advanced Energy Harvesting Piezoelectric Energy Harvesters," Ferroelectrics, vol. 599, no.
Journal, vol. 17, no. 1, pp. 34–47, 2024. 1, pp. 58–71, 2022.

[8] M. Zhang, et al., "Energy Harvesting Systems: A Review of [17] Z. J. Chew and M. Zhu, "Adaptive Self-Configurable
Piezoelectric and Triboelectric Technologies," Renewable and Rectifier for Extended Operating Range of Piezoelectric Energy
Sustainable Energy Reviews, vol. 90, pp. 463–473, 2016. Harvesting," IEEE Transactions on Industrial Electronics, vol.
67, no. 4, pp. 3267–3276, 2019.

[9] J. Li and H. Zhang, "Modeling and Optimization of

Piezoelectric Energy Harvesting for Urban Applications," [18] E. Lefeuvre, A. Badel, C. Richard, and D. Guyomar,
Energy Conversion and Management, vol. 125, pp. 335–347, "Piezoelectric Energy Harvesting Device Optimization by
2016. Synchronous Electric Charge Extraction," Journal of Intelligent
Material Systems and Structures, vol. 16, no. 10, pp. 865–876,
2005.
[10] H. Singh and P. Verma, "Towards Smart Cities:
Piezoelectric Energy Harvesting with Machine Learning for
Urban Power Grids," Sustainable Cities and Society, vol. 61, pp. [19] Y. Zhao, L. Zhang, and Q. Liu, "Optimizing Footstep
102288, 2020. Energy Harvesting Tiles Using Machine Learning-Based
Control," Smart Materials and Structures, vol. 29, no. 11, pp.
115012, 2020.
[11] K. K. Selim, I. H. Smaili, H. M. Yehia, M. M. R. Ahmed,
and D. A. Saleeb, "Piezoelectric Sensors Pressed by Human
Footsteps for Energy Harvesting," Energies, vol. 17, no. 10, pp. [20] A. Das, M. R. Islam, and F. Ahmed, "A Review on Recent
Advancements in Footstep Power Generation Systems," Journal
2297, 2024.
of Renewable and Smart Energy, vol. 10, no. 2, pp. 88–102,
2022.
[12] D. Ma, G. Lan, W. Xu, M. Hassan, and W. Hu,
"Simultaneous Energy Harvesting and Gait Recognition Using
Piezoelectric Energy Harvester," arXiv preprint
arXiv:2009.02752, 2020.

[13] L. Lenka, M. H. C., R. D. S., and R. B. N., "ML Assisted

Foot Step Power Generation Using Piezoelectric Sensors,"
International Journal of Advanced Research in Science,
Communication and Technology, vol. 4, no. 5, pp. 45–52, 2024.

96
PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

97
PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

98
PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

99
 PREDICTIVE ANALYSIS OF POWER GENERATION FROM FOOTSTEPS USING MACHINE LEARNING

Appendix B: Glossary

AI - Artificial Intelligence
ML - Machine Learning
IoT - Internet of Things
RF - Random Forest
LR - Linear Regression
GUI - Graphical User Interface
DB - Database
CSV - Comma Separated Values
CPU - Central Processing Unit
kWh - Kilowatt-Hour (unit of energy)
mAh - Milliampere-Hour (unit of battery capacity)
V. - Voltage
A - Ampere (unit of electric current)
J - Joule (unit of energy)
Pa - Pascal (unit of pressure)
UML - Unified Modeling Language
MA - Moving Average (used for voltage smoothing)
API - Application Programming Interface
RMS - Root Mean Square (for signal measurements)
DAQ - Data Acquisition
HTTP - HyperText Transfer Protocol

100